By 5:21pm Eastern, Fable 5 and Mythos 5 were gone. Not throttled. Switched off, for every customer, everywhere.

By the end of this issue you’ll have his playbook, plus my own verified setup, so you can run capable AI on your own machine, for free, with no company and no government able to flip the switch.

No YouTube video from me this week. I'm writing this from Cannes, coffee in hand and a beautiful view of the bay in front of me. I’ll be back in July.

If you’re new here, welcome. If you’ve been reading for a while, thank you. Your subscription is what makes issues like this possible.

MarTech AI is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

You don’t own the model

This was the first time the United States has ever issued an export-control directive for access to an AI model. The order barred foreign nationals from using Fable 5 and Mythos 5, including Anthropic’s own staff who aren’t US citizens. To comply, Anthropic had to pull both models for everyone.

I’m not here to argue whether the order was right. Plenty of people are doing that already. The lesson sits underneath the politics, and it applies to you whichever model you pay for.

You don’t own the model. You rent access to it.

Greg Isenberg put it best, and this whole issue is built on the point he made on LinkedIn this week. His framing: don’t build your entire workflow on something that can disappear with one letter. Own part of your stack.

He’s seen this film before. He once ran a startup whose biggest traffic source was organic Facebook. The algorithm changed overnight and he lost 99% of it. Anyone who built on rented reach got the same lesson. Email lists exist because people learned it the hard way.

You can already see it happening, and not just with hobbyists. PewDiePie, one of the biggest creators on the planet at 100 million plus subscribers, just went fully local.

He built a 10-GPU rig at home, ran a stack of open models against each other to pick the best answer, and released his own self-hosted AI workspace called Odysseus (https://github.com/pewdiepie-archdaemon/odysseus) with his own model in the works.

He owns the whole thing rather than renting it. Odysseus is an agent harness, the same idea I broke down in.

When the biggest channel on earth builds its own, you know where this is going.

So here’s how to setup your first local LLM this weekend.

Local models are insurance. You keep using frontier models for the heavy lifting. You keep a local fallback for the day access vanishes, or you need privacy, or the wifi drops on a train and you still want to work.

I call the setup the Local Stack. Eight steps, in order.

1. Pick your runtime (Ollama or LM Studio).
2. Match the model to your hardware.
3. Pick the right model for the job.
4. Shrink it with quantisation.
5. Point at your local model.
6. Mind the context window.
7. Give the model tools.
8. Fine-tune later.

What you need

• A laptop (a Mac with 16GB or more of memory is ideal)

• Ollama or LM Studio, both free

• One model downloaded to start (a small one)

• Twenty minutes

The first five steps get you running. The last three make you good at it.

If you want the secret sauce behind how I generate GIF infographics like this, apply here.

Step one → Pick your runtime

A runtime is the program that runs a model on your own computer. It’s the engine. No runtime = no local AI.

You have two free options. Ollama is the one I use. Download it from ollama.com, install it like any app, and it quietly runs in the background.

LM Studio is the other, with a friendlier window if you’d rather click than type.

Once Ollama is installed it serves models at a local address on your machine, localhost:11434. You don’t need to understand that line to use it. Just know it’s the door everything else knocks on later.

To download your first model, you run one line:

ollama pull [model]

That pulls the model onto your hard drive. From then on it runs offline.

Tip: Start with one small model before you download a 40GB monster you’ll never use. You can always pull more.

Step two → Match the model to your hardware

A model’s size is measured in billions of parameters. You’ll see numbers like 7B, 27B, 70B. Bigger is smarter, and bigger needs more memory.

Here’s the rough rule:

• A 7B model runs on almost any laptop

• A 27B to 32B model wants 32GB or more of memory

• A 70B model needs serious hardware, like a maxed-out Mac Studio

If a model is too big for your memory, it spills over and runs three to five times slower. So match the model to the machine you actually own, not the one you wish you had.

Use this step when: you’re about to download a model and you’re not sure your laptop can handle it. Check the size first.

Step three → Pick the right model for the job

Different open models are good at different things. These four cover almost everything a marketer or operator needs:

• Qwen 3 is the best all-rounder for most tasks.

• DeepSeek is the one for reasoning and coding.

• Gemma 3 is tiny enough to run on a phone.

• Llama has the biggest community.

You’re not locked in. Pull two or three, test the same prompt across them, keep the one that answers best for your work.

Tip: The “best” model on a benchmark is rarely the best model for you. The right model is the one that runs smoothly on your machine and gives you usable answers. A fast 7B you actually use beats a 70B you never load.

Step four → Shrink it with quantisation

This is the one concept that changes everything, and almost no one outside the field knows it.

Quantisation shrinks a model so it runs on weaker hardware, with barely any drop in quality. When you browse models you’ll see versions labelled Q4 or Q5. Those are the shrunk ones. Q4 keeps roughly 95% of the quality at a fraction of the memory.

This is how a model that “needs” a server ends up running fine on your laptop. Look for the Q4 version by default. Learn this word and the whole thing opens up.

Step five → Point Claude Code at your local model

This is the payoff, and it’s the part I verified myself across all those versions.

Claude Code is the terminal tool a lot of us now build with. By default it talks to Anthropic’s cloud. You can point it at a model running on your own machine instead.

You don't need to touch a config file. With Ollama installed, one line does the whole thing:

ollama launch claude

A menu pops up. Arrow-key to your local model, press Enter. Behind the scenes Ollama points Claude Code at your machine instead of the cloud, so your requests never leave your laptop. Now the tool runs on a local model. Free. Offline. Yours.

Tip: you’ll want a recent Ollama (v0.15 or later) for the launch command. Update first if ollama launch isn’t recognised.

Want more than Ollama’s one-liner? There’s a popular open-source proxy called free-claude-code (https://github.com/Alishahryar1/free-claude-code), 35k stars and counting, that points Claude Code at 17+ backends, local models or free API tiers, so you can swap which model sits behind Claude Code without touching a config file.

I won’t oversell it. A local 7B won’t match a frontier model on your hardest problems, and I’ll come back to that. But for a huge share of everyday work, a tuned local coder model is plenty. And on the Friday a model gets pulled, this is the difference between your workflow stopping and your workflow carrying on.

Use this step when: you want a fallback that survives an outage, a ban, or a flight with no wifi.

Step six → Mind the context window

The context window is how much the model can hold in its head at once. In the cloud you get a huge one for free. Locally, you pay for it in memory.

A bigger context window eats memory fast, and a model that was running smoothly will start to choke. So keep your sessions tight and your prompts lean. Clear the history when you move to a new task. This is the trade you make for running on your own machine, and once you feel it, you manage it without thinking.

Step seven → Give the model tools

A smaller local model with web search, file access and code execution beats a giant model with none.

The model is the engine. Tools are the wheels. Wire up the right ones and a modest local model handles real work, because it can look things up, read your files and run code rather than guessing from memory alone.

The capability gap between local and frontier closes fast the moment you add tools.

Tip: Before you chase a bigger model, give your current one tools. It’s usually the faster win.

Step eight → Fine-tune later

You don’t need this on day one. You should know it exists.

Fine-tuning means taking an open model and training it on your own data so it gets good at your specific work, your brand voice, your domain. It’s more accessible than it sounds, and it’s the reason owning an open model matters. You can shape it. You can’t do that with a model you only rent.

Park this until the first five steps are second nature. Then come back to it.

The honest limitation

Local models are not a replacement for frontier models. They’re insurance.

A local 7B won’t match a Fable-5-class model on your hardest reasoning, your trickiest code, your longest documents. The models that get close need real hardware, the kind most people don’t have at home yet. And context costs you memory every single session.

So this isn’t a breakup with the cloud. I still reach for the best frontier model for the heavy work. The point is to never again have my whole workflow depend on access I don’t control. Own one layer. Keep a local fallback ready. That’s the insurance policy, and the premium is twenty minutes this weekend.

The move this week is small. Install Ollama. Pull one 7B model. Run one prompt offline. That’s the whole policy bought.

The people who set this up before they need it will be the ones still working on the day the next letter lands.

Stay curious, stay human, and own the part of your stack nobody can switch off.

— Charlie

P.S. Hit reply and tell me one thing you’d want running locally if the cloud went dark tomorrow. I read every reply, and the best ones shape next week’s issue.

Ways we can work together

Become a Paid Subscriber: Get full access to every past issue of MarTech AI and support future ones.

Claude Content Engine: The Claude engine behind 350k+ followers (across socials) and 100 million impressions a year.

AI Second Brain for LinkedIn: Built on my methodology and frameworks. Learns your voice and builds your content strategy. Waitlist open.

Everything below still works, just pick Opus 4.8 instead of Fable 5 when you switch models.

Claude Code is brilliant at building things that work and terrible at making them look good.

Left alone, it gives you the same site as everyone else, with the purple gradient, the Inter font and the identical bento cards.

I built charliehills.ai anyway, in one evening, with three markdown files and a workflow that strips the AI look out completely.

The trick is not one-shotting “build me a website”.

By the end of this newsletter, you’ll have the exact nine steps and every prompt I used.

If you are new here, welcome. If you have been reading for a while, thank you. Your subscription is what makes issues like this possible.

I filmed the whole build, every step on screen. Watch it alongside this issue:

Step one → Write the facts file

Then create the facts file. I made mine in Claude Chat rather than Claude Code, because Chat already knows me from months of conversations.

Paste this into Claude Chat:

You know me from our conversations, but do not assume this site is about me. I'm building a website and need a file called CONTEXT.md holding the facts it should be built from. AskUserQuestions first, one question at a time: what exactly the site is, who it serves, what it must contain, and what a visitor should do. Keep asking until you could brief a stranger on the build. Then show me the fact list you plan to include and wait for my yes. Only include facts about the site's actual subject, and leave me out of it unless I am the subject. Real facts from my answers and from what you know, nothing invented. Give it to me as a downloadable markdown file.

Download the file, drop it into the project folder, and correct anything it got wrong. If Chat doesn’t know you yet, tell Claude Code to create CONTEXT.md and interview you for the facts instead.

Mine holds:

• Who I am and what I do

• Every number and proof point

• My story, my frameworks and my voice rules

Step two → Let Claude interview you

Make a new folder, open your terminal there, and type claude to start Claude Code.

Switch to Fable 5 with the /model command.

You don’t write the website copy.

You answer questions and Claude drafts it.

Paste this:

Read CONTEXT.md. Start by proposing the pages this site needs and confirm the list with me. Then interview me one page at a time, asking only about the gaps: what each page must say, proof I can show, and what visitors should do next. Also ask what searches each page should rank for. Then draft every page's copy into one file called COPY.md in the project root and show me before you build anything. Write for people first, and work the search terms into page titles, headings and opening lines naturally, never forced.

Step three → Approve every word before any code

The interview produces COPY.md: every headline, every section, title tags, meta descriptions, an SEO map per page.

Tell Claude what to change and it edits the file.

Step four → Merge your brand with sites you love

DESIGN.md is your site’s rulebook. Your job is gathering the ingredients, and Claude does the merging.

1. Find the looks you love. I browsed Refero Styles and picked the design systems of sites whose look made me jealous: like Air and Apple. Aim for around five:

2. Download their DESIGN.md files. Open each style, click its DESIGN.md tab, and hit the .md download.

3. Add your brand kit. Mine is a PDF with my colours and font. If you have no brand kit, a text file with two colours and one font is enough.

4. Drag everything into Claude Code and paste:

I've dropped in design files from sites I love, plus my brand kit. Merge them into one DESIGN.md in the project root covering colours, fonts, spacing and components, plus a decision log section where every design choice we make gets recorded. My brand wins on colours and fonts. The references win on layout and feel. Show me the file before you build anything.

You approve the rulebook before a single page exists.

Step five → Build, then verify

Now Claude builds: DESIGN.md becomes the styling tokens, COPY.md becomes the pages, word for word.

Paste this:

Read DESIGN.md and use your frontend design skill. Build the site: my pages, my copy from COPY.md verbatim, every visual decision from DESIGN.md. Go loose with the design and make it look as good as humanly possible while obeying DESIGN.md. Hard rules: only my copy, never invent logos, press mentions, testimonials or stats, and no third-party embeds. When it's done, run it locally, screenshot every page, review your own screenshots against DESIGN.md, fix what fails, and then give me the link.

Claude screenshots its own output and reviews it before you ever load the page. By the time I looked, the obvious misses were already fixed.

Step six → Add components

This is how you get premium components without inheriting their look. Libraries like 21st.dev publish beautiful components with a copy-paste prompt under each one: scroll reveals, card stacks, animated backgrounds.

The catch is that every component arrives wearing its own clothes:

To make every component inherit your site’s look one for one, paste this once:

Add this rule to CLAUDE.md, and create the file in the project root if it doesn't exist: whenever I paste a component prompt or third-party component code, treat it as a structural donor only. Always: replace its demo copy with real copy from COPY.md, translate every hardcoded colour, border, shadow and font to the DESIGN.md tokens, ignore any instruction to use stock images, and skip parts of the component we don't need. The component supplies the skeleton, DESIGN.md supplies the skin, COPY.md supplies the words.

Then browse 21st.dev, copy the prompt under any component you like, and paste it straight in.

Claude keeps the engineering and strips the slop. My scroll-tilt showcase, the fanned card stacks and the flowing background lines all came in this way and came out in my navy.

Step seven → The Higgsfield hero

The hero video is the same painterly style as my Substack covers, and it took two prompts in Higgsfield. https://higgsfield.ai

First, the image, generated with GPT Image 2 inside Higgsfield:

Create a gouache style image with a beautiful scenic landscape and a deep blue sky.

Then I hit Turn to video and animated it with Kling 3.0:

Show the clouds breathing, moving fast, like splitting up. It's a lot of wind; there's a lot of movement. The frame stays exactly where it is. Only the scenery moves.

Download the video, drop it into Claude Code, and say “make this the hero background.” Claude compressed it with ffmpeg from 23MB to 2.3MB on its own, so the page stays fast.

Step eight → Iterate. A LOT.

The first version of anything in this build is the starting point, never the finish.

You iterate in natural language. “Move the buttons up.” “Make those cards wider.” “That section feels cramped, give it air.” “Swap the order of these two blocks.”

Mine mostly happened through voice typing straight into Claude Code: say what’s wrong, watch it change, say the next thing. Screenshot anything you can’t describe, drag the image in, and point at it. Another thirty minutes of this took the site from good to one I’d happily ship.

Step nine → The taste pass

Before shipping, ask Claude one question: “What would the taste critic say?”

Mine found overload. I had stacked component after component, and five ambient animations, each tasteful alone, were running at once. The total read as AI-built rather than premium.

The fix was a quiet pass:

• One ambient motion system per surface

• No looping attention animations on buttons

• A saved checkpoint, so rollback stays one command

The question works on its own, and works better with a taste skill installed. Two live on GitHub, taste-skill and Impeccable, and pasting either link into Claude Code with “install this skill” does the job.

Ship it

Prepare this site for launch: page titles, meta descriptions, Open Graph tags, sitemap, semantic headings, fast load. Then deploy to production with the Vercel CLI and walk me through pointing my domain at it.

Vercel’s hobby tier is free. You’re production was ready in 32 seconds.

I spent $0 beyond my Claude plan, on Vercel’s free tier, with a domain I already owned.

The full stack in one list

• Claude Code (Fable 5): https://claude.com/product/claude-code

• Refero Styles (DESIGN.md library): https://styles.refero.design

• 21st.dev (component prompts): https://21st.dev

• Aceternity (component prompts): https://ui.aceternity.com

• MotionSites (hero and animation prompts): https://motionsites.ai

• taste-skill (the anti-slop pass): https://github.com/Leonxlnx/taste-skill

• Impeccable (spacing and hierarchy): https://impeccable.style

• Higgsfield (hero video): https://higgsfield.ai

• Vercel (free hosting): https://vercel.com

Try it on your own domain

Websites used to be hard. Pick a domain you already own, write your facts file, and run the nine steps in order. The site at the end is yours: your words, your colours, your proof, no stock anything.

Send me what you build, I read every reply.

Stay curious, stay human, and stop shipping the default.

— Charlie

Where you can meet me this month

VivaTech, Paris: Wednesday the 17th to Friday the 19th of June. I’m one of this year’s ambassadors. Use CHARLIE20 for 20% off any pass: https://vivatech.com/get-your-pass

Cannes Lions: I’ll be in Cannes from Friday the 19th to Sunday the 28th. If you’re heading down, I’d love to meet you at this year’s event.

The AI Beach Party: Monday the 29th, where I’m speaking: https://aibeachparty.co.uk

LinkedIn Personal Branding MASTERMIND, London: I’m teaching the AI visuals session, building scroll-stopping infographics live with no design skills needed. Tickets: https://luma.com/personalbrandingmastermind

Ways we can work together

Become a Paid Subscriber: Get full access to every past issue of MarTech AI and support future ones.

Claude Content Engine: The Claude engine behind 350k+ followers (across socials) and 100 million impressions a year.

AI Second Brain for LinkedIn: Built on my methodology and frameworks. Learns your voice and builds your content strategy. Waitlist open.

Level 1 runs from one command.
Level 2 runs it on a schedule you set.
Level 3 runs everything else.

Most people never get past zero.

Today I’m taking you up the whole ladder, from Level 1 to Level 3: what each one is, how to set it up, and when to stop climbing.

Read more

To the 9% who said no or wanted a different topic, sorry.

What do you want to see in next week's newsletter? Vote below. If it's not listed, reply to this email and tell me what you want featured instead.

For everyone else, here it is.

If you've been holding off on YouTube because editing is too much hassle, this is for you. The pipeline took my channel from zero to 3k subscribers and 350k+ views.

Read more

I help creators and marketers get more done with AI.

You probably landed here because I sent you a resource. It lives inside this vault, along with everything else I’ve made. MarTech AI is where I share the actual prompts, skills and workflows I use, every week, for free.

New here? Subscribe (free) and the full vault unlocks: 100+ prompts, Claude skills, guides, cheat sheets and tool stacks. Updated most days.

Important: access is not instant. After you subscribe, a welcome email lands in your inbox within a few minutes. That email holds your vault link. So please:

1. Wait a few minutes for it to arrive

2. Check your promotions tab

3. Check your spam folder

Once you’re in:

1. Confirm your subscription

2. Reply with a quick hello

3. Move the email to your primary inbox

Once you find the URL, bookmark the link and you will never need to search again.

Subscribe free and it’s all yours 👉 https://charliehills.substack.com/subscribe

Already subscribed? You've already got the vault link in a previous giveaway email or in your welcome email. Find it, bookmark it, and you'll never need to search again.

Stay curious, stay human and keep creating.

– Charlie

So I did what I typically do and wrote a newsletter about it.

Today, I define the term, show you what one is, and walk you through how to build each piece in Claude Code or Cowork.

The term was first coined in February of this year. OpenAI and Anthropic expanded on it within weeks. The phrase is now everywhere in the AI world.

It sounds new. It is not…

If you run Claude Code (or even Cowork), you’ve been operating one for months. You just called it something else.

Prompt engineering (2022 to 2024). Get the wording right.

Context engineering (2024 to 2025). Andrej Karpathy’s term. Get the right information in front of the model every time.

Harness engineering (2026). Mitchell Hashimoto’s term. Everything context engineering does, plus tools, memory, guardrails, and specialists.

Read more

You met Claude Code. This week, we build on top of it.

Agents are easier than they look.

Below are the 6 steps that took me from prompts to a working agent team.

If you are new here, welcome. If you have been reading for a while, thank you. Your subscription is what makes issues like this possible.

Read more

I called it “Beginner to Advanced”.

I was the beginner…

Six weeks later, I am still not the advanced one. But Claude Code now runs about 80% of my content business.

Claude Code is genuinely easy. The CLI (Command Line Interface) is just another text box. You type what you want. Claude does it. That is how it works.

The fear is the only thing in your way.

Below are the 6 features that took me from terrified to fluent.

You will finish this email going “oh, I can do that too”.

If you are new here, welcome. If you have been reading for a while, thank you. Your subscription is what makes issues like this possible.

Read more

Today I’m giving you my honest tier list of every single feature, so you know:

☑ What to install this week
☑ What to delete from your habit

Anthropic ships a new feature almost every week.

Every launch post tells you the same things:

1. The new thing is essential

2. You should switch to it immediately

3. You should rebuild your whole workflow around it

The truth is much messier than that.

Some features run my entire business. Whereas others I’ve never opened again after trying them.

If you are new here, welcome. If you have been reading for a while, thank you. Your subscription is what makes issues like this possible.

Read more

Today, I'm giving you a full insight into the simple system behind all of it.

I’m just getting started on YouTube. If you want to check that out, I recorded a video walking through this simple content process:

Read more

Andrej Karpathy posted a GitHub gist on 3rd April. Within 48 hours, his X thread about it hit 16 million views.

It was 1,500 words with no code or repo. An idea file describing how to build a personal knowledge base with an LLM and a folder of markdown.

Read more

I one-shot a working app prototype.
I generated a fully designed pitch deck.
I handed a product demo off to Claude Code.

By the end of this newsletter, you will know the exact workflow to turn a rough idea into a working on-brand prototype in under ten minutes, plus the two output types, the four refinement tools, and the one setup step most people will skip.

If you are new here, welcome. If you have been reading for a while, thank you. Your subscription is what makes issues like this possible.

I also recorded the full video version of this newsletter. If you would rather watch than read, everything is in there:

Design used to mean juggling Figma, Canva, Lovable, and Claude Code. Every handoff broke something.

Claude Design collapses the stack into one interface.

It gives you a place to prompt, edit, comment, and export.

You need a brand kit and a clear brief. Nothing else.

The 4 starting points

Claude Design opens with four templates. Pick the one matching your output.

1. Prototype. Pick this for app UIs, dashboards, SaaS editors.

2. Pitch Deck. Pick this for slides, investor decks, client presentations.

3. From Template. Pick this for interactive product pages and landing demos.

4. Design System. Pick this when you want Claude to learn your brand before anything else.

Each runs the same two-step flow. Upload your brand kit. Write a brief. Everything else happens inside the editor.

The 5-step workflow

Before you start: set up your design system

Build your design system first. Every project you run after inherits it. Skip this step and your assets drift off-brand across three prompts.

What Claude does during setup:

☑︎ Reads your codebase
☑︎ Reads your existing design files
☑︎ Builds a shared design system for your team
☑︎ Auto-applies branding and components to every project

Your prototypes come out on-brand. Your pitch decks match. Your landing pages pull the same spacing rules.

Design system first, assets second.

With your system in place, here’s the 5-step workflow:

Step 1: Name your project and pick a type

Name your project first. Claude uses the name as context across everything inside.

Then pick Prototype, Slide deck or From template. If you are in Prototype, decide between Wireframe and High-fidelity before you write a word.

Tip: Start every new product idea in Wireframe. Lock the layout. Duplicate into High-fidelity once the structure is right. You save tokens and avoid rebuilding polished designs around a layout change.

Step 2: Upload your inputs

Claude Design accepts six input types:

1. Images

2. Text prompts

3. A codebase link

4. DOCX, PPTX, XLSX uploads

5. A web capture from your live site

6. Your saved design system

No design system yet? Drop your brand kit into the prompt. Include your logo, colour palette, typography, and a few reference screenshots of designs you admire.

Tip: The “Grab web element” capture tool is the sleeper feature. Point Claude at your live website and the prototype inherits the look of the real product.

Step 3: Write a specific brief

Vague briefs produce generic output. Specific briefs produce usable prototypes.

Weak brief:

“Design an app.”

Strong brief:

“Build an interactive prototype of the Vislo editor. Include a left sidebar for canvas selection, a main editing area, a top toolbar with export and share buttons, and a right-hand panel for layer controls. Match the attached brand kit.”

The difference is layout detail. Claude needs to know the regions, components, and actions you expect on screen.

Tip: Attach a reference UI screenshot for the overall feel. Visual cues travel faster than paragraphs of description.

The prompt I used to build this:

Build an interactive prototype of the Vislo app editor. First screen is My Designs, a grid of recent infographic cards with a prominent New Design button. Clicking New Design opens a prompt input with four suggested templates below: Stat Sheet, Timeline, Comparison, Process Flow. After a prompt submission, animate the transition into the editor view: a centre canvas with the infographic, a left layers panel, a right properties panel, and a top bar with Undo, Redo, Share, Export. Make the template picker, the prompt input, and the Export menu tappable. Include a hover state on cards and a loading shimmer on the canvas during generation.

Step 4: Refine your design live

Claude Design gives you four ways to edit without regenerating:

1. Inline comments on any element

2. Direct text edits inside the design

3. Custom sliders for spacing, colour, and layout

4. Apply a change across the full design

Click an element, drop a comment in plain English, and Claude updates only that element.

From my session: “Change this to a prompt bar instead of a prompt box.”

The live prototype updated and the rest of the design stayed intact.

Tip: Use inline comments for single-element tweaks. Use Apply across for anything global. Mixing them saves tokens and keeps the parts you liked.

Step 5: Export or hand off to Claude Code

Export options include PDF, PPTX, Canva, and standalone HTML. You also save to a folder or share via an organisation-scoped internal URL.

For production UI work, the handoff to Claude Code is where Claude Design earns its keep.

Claude packages the design into a handoff bundle. One instruction inside Claude Code picks it up. Spin up a local host and iterate from there.

The alternative is building inside Claude Code from scratch, adding design plugins, installing skills, and hacking the output until the UI looks presentable. Claude Design removes the first 80% of that work.

How to build a slide deck from template

The Slide deck flow follows the same project setup. Output is HTML-coded slides rather than a web app.

My Vislo pitch deck took around 8 minutes to generate. Longer than a Canva template, but the output is stronger, with custom layouts, proper logo use, and on-brand typography.

The prompt I used to build this:

Read more

I’ve done this hundreds of times. It works. But there are three more levels above this. I've spent the last few weeks testing all three.

The short version: AI design is moving from accepting outputs to owning them.

If you are new here, welcome. If you have been reading for a while, thank you. Your subscription is what makes issues like this possible.

I also recorded the full video version of this newsletter. If you would rather watch than read, everything is in there.

Level 1: Gemini (prompt to image)

This is where most people start.

And honestly, where most people stop.

You describe what you want. Gemini builds it.

“Create an infographic showing 5 ways to use AI for email marketing. Handwritten notebook style, blue and white colour scheme.”

Thirty seconds later you have a visual. For single-use infographics and social posts, this is more than enough.

The handwritten style was a total pattern interrupt on LinkedIn. Then everybody started doing it. Now it’s gone from a pattern interrupt to a pattern. They all look the same.

The limitation is you get what you get.

If the output is 90% right but the headline font is wrong or the spacing feels off, your only option is to regenerate and hope.

I’ve done this dance more times than I’d like to admit. Generate, squint, regenerate, squint harder, settle.

I’ve posted dozens of Gemini-generated infographics on LinkedIn. They perform. But every single one is a standalone creation with no memory or brand consistency between sessions.

Use Gemini when you need a quick visual for a single post and you don’t need it to match existing brand assets.

Level 2: Claude Chat (HTML/CSS in the chat interface)

You do this in Claude's chat interface or Cowork too. Instead of generating an image like Gemini, Claude writes the visual as HTML and CSS code.

I prompted Claude with “create an HTML CSS infographic on Claude’s AI ecosystem” and it generated a decent visual. Not bad for a first attempt.

The problems: it’s not optimised for social media dimensions (1080 x 1350), it’s not on brand, and the iteration process burns through your usage incredibly fast. Even on the 10x Max plan I hit my limits.

I wrote a full guide on how to manage your Claude usage. Here's the infographic version (which was created by Level 3).

You upload your brand kit, Claude rewrites the code to match your colours and typography. But every change is another generation. Another chunk of your usage gone.

The output is more structured than Gemini, but Gemini is more creative and better for illustrations. This is a trade-off, not an upgrade. You’re swapping visual creativity for layout control.

This is where most people will land if they try this today. It works. But the cost per infographic is high and the process is slow.

Use Claude Chat when you want more structured layouts than Gemini but you’re not ready for Claude Code.

Level 3: Claude Code (code-generated design)

This is where things get interesting. And where I’ve been living for the last few weeks.

I made a full breakdown of the 9-step process:

Now, let me walk you through what's actually happening behind this.

Instead of describing an image and hoping for the best, you give Claude Code your brand rules, your hex colours, your font preferences, and your layout specifications.

It generates the visual as HTML and CSS.

You get a file you own. Every pixel is editable.

Two posts I made with Claude Code recently:

“54 Claude Code resources” → 171,915 impressions.

“Claude Code has 93 commands” → 123,024 impressions.

Both used infographic visuals generated entirely through code. If you’ve been following my LinkedIn lately, you’ve been looking at Claude Code output.

I bake the brand rules directly into the system, so every infographic follows the same colour palette, typography, and spacing logic. My audience now recognises the style before they read the content.

The consistency compounds. Each new piece reinforces the one before it.

How I set this up

Read more

Everything in this newsletter is tested. I ran every workflow, hit every error, and iterated until it worked.

This took me over 10 hours to put together. I covered every feature across Chat, Cowork, and Code. Every how-to. Every tip. Every workflow I actually use. Nothing I have not personally tested.

If you are new here, welcome. If you have been reading for a while, thank you. Your subscription is what makes issues like this possible.

I also recorded the full video version of this newsletter. If you would rather watch than read, everything is in there. Chat, Cowork, and Code, level by level.

Anthropic shipped 74 product releases in 52 days.

Revenue went from $9 billion to $19 billion in three months. Claude Code alone generates $2.5 billion a year. 8 of the Fortune 10 are customers. Software stocks crashed on launch day. Twice.

And most people are still using Claude like a chatbot.

Claude is three products. Not one.

Chat is where you think. Cowork is where you produce. Code is where you automate.

Each product has its own set of features. Most people know Chat. Some have opened Cowork. Almost nobody outside of developers has touched Code.

Today I am walking through all three. Every feature. How to turn it on. How I use it.

Claude Chat

This is where everyone starts.

1. Go to claude.ai
2. Open the app
3. Type…

But Chat has 12 built-in tools. Most people use 3.

Here is the full list:

Models + Extended Thinking

Opus 4.6 is the most capable model. 1M token context window. Use it for strategy, legal analysis, financial modelling, and anything with multiple variables.

Sonnet 4.6 handles 90% of your daily work. Faster. Cheaper. I use this for everything unless I have a reason not to.

Haiku 4.5 is the lightweight option. Fast and cheap. Good for quick answers and bulk tasks where depth does not matter.

My rule: start with Sonnet. Switch to Opus when the task demands it. Most people leave Opus on permanently and burn through their usage limit by Wednesday.

Below the model selector you will see Extended Thinking. Toggle it on. Claude reasons through your problem step by step before responding. Shows its working. Breaks complex questions into parts.

I never prompt Opus without Extended Thinking on. The difference is like asking someone to answer immediately versus giving them 5 minutes to think. Turn on Extended Thinking and Web Search at the same time. Two clicks that change everything.

How to access: model dropdown at the top of any conversation.

Memory

Claude remembers you across conversations. Free for all users since March 2026.

Two settings to enable in Settings → Capabilities:

Search and reference chats → Claude searches your past conversations for relevant context.

Generate memory from chat history → Claude remembers details about you across all chats and Projects.

Turn both on. This is what makes Claude feel like it knows you.

If you are coming from ChatGPT, you can import your existing memory. Go to Settings → Capabilities → Memory and click Start Import. Claude gives you a prompt to fetch your memory from your other account. I wrote a full guide on how to do this, linked below.

Clean up your memory regularly.

Claude sometimes misinterprets things and adds random details about you. Old projects you are no longer working on, incorrect assumptions about your role, outdated preferences. Go into your memory, read through it, and remove anything that is wrong or stale.

You can also actively tell Claude what to remember. Say “remember that I now work at [company]” or “remember that I prefer responses in British English with short paragraphs” directly in any chat. Claude saves it to memory, as long as you have the settings enabled.

Tip: Treat your memory like a living document. Review it every few weeks. Remove what is outdated. Add what is missing. The cleaner your memory, the less time you spend correcting Claude in every conversation.

Web Search

Claude searches the web mid-conversation and pulls live data into its response.

Two ways to trigger it:

→ Toggle it on in Settings (stays on for every chat)
→ Or type “search the web for...” in your prompt (one-off)

I keep mine toggled off by default to save tokens. When I need it, I type “search the web” directly in my message. That’s faster than enabling it, and I only burn the tokens when the task needs live data.

Tip: pair Web Search with Extended Thinking for your most complex questions. Claude pulls live data AND reasons through it step by step. Two toggles that change the quality of every response.

Deep Research

This is different from Web Search. Deep Research runs a full multi-source investigation. Claude searches dozens of sources, cross-references them, and returns a structured report.

How to use:

→ Click the Research button next to the message bar
→ Enter your topic
→ Wait for the full report (takes a few minutes)

I sometimes use it before every newsletter. I give it a topic, it comes back with data points, competitor examples, and source links.

The key: be specific.

“Research LinkedIn growth strategies” → generic output.

“Research what strategies LinkedIn creators with 100K+ followers used in Q1 2026, with specific examples and engagement data” → something I can publish from.

Tip: Deep Research burns through tokens fast. Use it for the research phase, then switch to a normal chat to write from the findings. Do not research and write in the same thread.

File Upload

Drag PDFs, images, spreadsheets, or documents directly into the chat. Claude reads the full content.

How to use:

→ Click the paperclip icon
→ Or drag files directly into the chat window
→ Works with PDFs, CSVs, images, Word docs, and more

Upload 2 or 3 documents at once and ask Claude to compare them. “Here are three competitor landing pages. What patterns do you see across all three?”

Tip: if you have a file on Google Drive, click the “From Drive” button instead of downloading and re-uploading. Claude pulls it directly from your connected Drive.

Artifacts

Claude builds interactive files, dashboards, calculators, and tools inside the chat.

How to use:

→ Ask Claude to build something
→ It creates the Artifact automatically
→ Use, edit, and download directly

Ask for what you would normally build in a spreadsheet or Canva. Budget trackers, content planners, data visualisations, landing page mockups.

Ask for changes after it builds. “Make it dark mode.” “Add a column.” “Change the chart type.” Artifacts are live. You iterate inside the chat.

Tip: say “make this an Artifact” if Claude gives you code in the chat instead of building it. Sometimes it defaults to showing code. Telling it to make an Artifact forces it to render the live, interactive version.

Interactive Charts

Added March 12. Claude now creates custom charts, diagrams, and visualisations inline in its responses. These are interactive, not static images.

How to trigger:

→ Ask Claude to “create a chart” or “visualise this data”
→ Claude renders it inline in the conversation
→ You interact with it directly in the chat

This is different from Analysis Tool. Analysis Tool runs code on your data. Interactive charts are visual outputs Claude builds from any information, not files only. Ask Claude to “create an org chart for my team” or “visualise this process as a flowchart” and it renders it live.

Tip: ask Claude to “make this interactive” after it generates a chart. You get hover states, zoom, and clickable elements instead of a flat image.

Interactive Questions

Claude presents clickable options instead of asking questions in plain text. Single select, multi-select, and drag-to-rank.

You have seen this if Claude has ever asked you to choose between options and given you buttons instead of a text list.

Two ways to trigger it:

→ Claude decides to use it automatically
→ Or type “AskUserQuestion:” in your prompt to force it

“Which format do you want?” → clickable buttons.
“Rank these in order of priority” → drag and drop.
“Which of these apply?” → multi-select checkboxes.

Tip: if Claude asks you a question in plain text and you would prefer clickable options, say “give me options to choose from.” Claude switches to the interactive format. Works the other way too. If you want to type a freeform answer instead of picking from options, ignore the buttons and type your response.

Projects

Save files and instructions once. Every new chat inside a Project starts with full context. Claude reads everything before you type a word.

How to set up:

→ Click “Create Project” in the left sidebar
→ Upload your key files and write custom instructions
→ Start new chats inside the Project

I run 12 active Projects. One per client, one for this newsletter, one for LinkedIn content. Each has a custom instruction, 3 to 5 reference files, and zero filler.

Most people do everything in one Project. I do the opposite.

I take a piece of content and move it through different Projects depending on what I need. A newsletter becomes a LinkedIn post, which becomes an infographic, which becomes an email subject line.

Each Project has its own custom instructions, tone rules, and reference files. The output is tailored every time without me re-explaining anything.

I switch between Projects throughout the day depending on the task.

Tip: if you upload the same PDF across multiple chats, Claude re-counts those tokens every time. Projects cache your files once. Upload once. Reference forever.

Paid subscribers: scroll to the bottom of this newsletter for my full newsletter Project instructions. Copy them directly into your own setup.

Skills

Projects are great when you need reference material and baked-in context. But when you need Claude to follow a very specific process the same way every time, that is where Skills come in.

Think of them like GPTs, but you recall them inside any conversation whenever you need them.

How to set up:

→ Go to Settings and enable Code Execution
→ Browse the pre-built Skills library
→ Install a Skill or create your own with a SKILL.md file

I have Skills for newsletter writing, carousel creation, and client reporting.

Here is one example. I write my newsletter, then ask Claude to generate a prompt for the Substack cover image. I built a Skill for this so every cover image follows the same style, dimensions, and brand guidelines. Consistency without me thinking about it.

You do not need to write Skills manually.

Claude has its own built-in Skill for creating Skills. You do not need to type out a SKILL.md file yourself.

Here is how it works in practice:

→ Explain the problem you are trying to solve
→ Provide reference material of what you want
→ Claude creates the Skill for you

Then you test it and refine it over time through natural language in chat. “Make the tone more casual.” “Add a step where you check for spelling.” “Always include a CTA at the end.” Claude updates the Skill based on your feedback.

That is exactly how I built my cover image Skill. I showed Claude a few examples of the style I wanted, described the dimensions and brand colours, and he wrote the full Skill. I have refined it a few times since, through conversation, and now it produces consistent results every time I run it.

Advanced: chaining multiple Skills inside a Project

You can set up a Project that calls multiple Skills as part of one process.

→ Create separate Skills for each stage (researcher, writer, reviewer)
→ Upload the empty SKILL.md files into your Project
→ Tell Claude in the Project instructions which Skill to call at each step

One prompt kicks off the entire chain. Claude researches, writes, and reviews, all following different Skills in sequence.

I do not personally follow this approach yet, but my co-founder at Linked Agency recently started using it for client work and it has been a standout for outsourcing repetitive deliverables where the process stays the same but the content changes each week.

Connectors (MCP)

Connect Slack, Google Drive, Notion, Gmail, and 50+ tools. Claude searches them mid-chat.

How to connect:

→ Click the connectors icon in any chat
→ Browse and authorise the tools you use
→ Claude pulls context without you uploading anything

I have Drive, Slack, and Notion connected. Last week Claude found a client brief from November I had forgotten existed. Pulled it from Drive in 3 seconds.

Tip: Connectors do not sync live. Claude searches on demand. Think of it as search, not a dashboard. Also, every connected tool adds to your token usage per message. Only connect the tools you use regularly.

Claude in Chrome

A browsing agent inside your Chrome browser. Reads pages, extracts information, completes web tasks.

How to install:

→ Go to claude.ai/download
→ Install the Chrome extension
→ Claude reads the page you are on and takes actions

Record a workflow and turn it into a shortcut.

This is the feature most people overlook. You walk through a task once and Claude remembers the steps.

How it works:

→ Click Record in the Chrome extension
→ Do the task manually while Claude watches
→ Claude captures every click, every navigation, every action
→ When you are done, it creates a reusable shortcut

That shortcut gets a name (like /manage-connectors), a saved prompt describing each step, and a starting URL. Next time you need to run that task, you type the shortcut and Claude handles it.

You can also schedule these shortcuts to run on a recurring basis. Weekly browser reports, daily form checks, regular data pulls from dashboards that have no API.

Tip: Start by recording a simple task you do every week in the browser. Something repetitive that takes 5 minutes. Once you see Claude replay it without you touching anything, you will find more things to record.

Claude Cowork

Cowork is where Claude moves from answering questions to doing work on your machine. It reads your files, creates real documents, and delivers finished outputs into your folders.

If Chat is the brain, Cowork is the hands.

You need the Claude Desktop app. Mac and Windows. Pro and Max plans.

→ Open Claude Desktop.
→ Click the Cowork tab.

That is your starting point.

File System Access

Cowork reads and writes to your actual folders. Word docs, spreadsheets, PDFs, presentations. Real files, not copies inside a chat window.

How to start:

→ Open Claude Desktop
→ Click the Cowork tab
→ Point it at a folder and give it a task

“Read everything in this folder and create a summary report.”
“Take these 5 client briefs and create a comparison table.”
“Rename and organise these 200 files by date and type.”

Tip: keep a .md file in every working folder that tells Claude who you are and what you need. Your role, your tone, your formatting rules. One file. Claude stops sounding generic.

Sub-agents

For complex tasks, Cowork automatically coordinates multiple sub-agents working in parallel. You do not set this up. Claude handles it.

What makes it useful beyond the speed is that you can steer mid-task. If Claude is running a workflow and you want to add something or redirect it, jump in and it folds your instruction into the current run. No need to start again.

Where it makes a difference:

→ Large tasks with independent components
→ Research across multiple sources at once
→ Anything sequential that takes too long

Tip: be specific about the output you want upfront. There is less back and forth when sub-agents are running in parallel, so the clarity needs to be in the original prompt.

Plugins

Skills are individual instruction files. Plugins bundle multiple Skills, connectors, slash commands, and sub-agents into one installable package for an entire role.

A sales Plugin connects to your CRM, adds prospecting workflows, and gives you /research and /outreach commands. A legal Plugin includes your NDA playbook, document management connector, and /triage-nda.

Cowork ships with 11 pre-built Plugins covering sales, finance, legal, marketing, HR, engineering, and more.

Four ways to add them:

Browse plugins → Pre-built library, one-click install.
Add marketplace → Your organisation’s private plugins (Team/Enterprise).
Upload plugin → Drop in one a colleague built.
Create with Claude → Describe what you want, Claude builds it.

How to access:

→ Click Customize in the left sidebar
→ Type / in any Cowork conversation to see your commands

Tip: Do not install all 11. Pick 1 or 2 that match your role. After installing, click Customize on any Plugin and Claude walks you through adjusting it to fit how you work.

Scheduled Tasks

Recurring prompts that run on autopilot. Morning inbox summaries, weekly reports, daily data pulls. Claude runs them at the time you set and delivers the results.

How to set up:

→ Go to Customize → Tasks
→ Write your prompt
→ Set the schedule and let it run

Start with one simple task. “Every Monday at 8am, summarise my unread Slack messages and flag anything urgent.” Once it works, add a second.

Build the habit of delegating one recurring task before trying to automate everything at once.

Where it gets interesting is when you bring Connectors into it.

Connect your Slack, Google Drive, or Notion, and your Scheduled Tasks stop being static prompts. They become live workflows that pull real data from your tools on a schedule, without you touching anything.

Here is one of mine.

Every morning at 7am, Claude runs my pre-configured Apify scrapers across Instagram, LinkedIn, X, TikTok, Threads, Facebook, and YouTube. It collects the 5 most recent posts from each account I track, including captions and engagement metrics, and then compiles everything into a daily content briefing. I open my laptop and it is already done.

One caveat: your computer needs to be awake for Scheduled Tasks to run. Set them for a time when your laptop is already open, your morning routine, a lunch break, the end of your working day.

Dispatch

Assign Claude a task from your phone. It runs on your desktop while you are away and messages you when the work is done. One continuous conversation across your phone and your computer.

How to set up:

→ Open Cowork on your desktop
→ Click Dispatch in the sidebar
→ Scan the QR code with the Claude mobile app
→ Send tasks from your phone. Come back to finished work.

I text Claude from my phone at 6am: “Compile yesterday’s LinkedIn metrics and draft this week’s client update.” By the time I sit down at my desk, the report is done.

Two things to know:

→ Your Mac has to be awake. Claude Desktop has to be open. I leave my Mac running overnight for this reason.

→ Dispatch is one continuous thread. You cannot start multiple threads. Everything lives in one conversation.

Tip: Dispatch gets more powerful when you combine it with Skills, Scheduled Tasks, and Connectors. Set up a morning briefing Skill, schedule it for 7am, check the results from your phone on the train. Your desktop does the work while you are not there.

Computer Use

Claude controls your screen, mouse, and keyboard. It opens apps, navigates browsers, fills in forms. macOS only. Released March 23.

How to enable:

→ Open Cowork
→ Go to Settings and enable Computer Use
→ Claude asks permission before accessing new apps

Claude reaches for the most precise tool first. If a Connector exists for Slack, it uses the Connector. If no API or integration exists, it falls back to controlling your screen. Computer Use is the last resort, not the first option.

It works best for apps and workflows that have no other integration. Internal dashboards, legacy software, anything without an API.

Tip: Computer Use runs outside the Cowork sandbox. When Claude uses your apps through screen control, it operates on your actual machine. Set up global instructions in Cowork like “never delete files without confirmation” and they carry over to Computer Use sessions.

Claude in Excel

An add-in that reads your entire workbook. Understands nested formulas, multi-tab dependencies, and provides cell-level citations for every explanation.

How to install:

→ Open Excel
→ Home → Add-ins
→ Search “Claude by Anthropic”
→ Open sidebar: Ctrl+Alt+C (Windows) or Ctrl+Option+C (Mac)

Ask Claude to trace errors first. “Why is cell B4 showing #REF? Trace the error.” Claude navigates across tabs, finds the broken reference, and fixes it while preserving dependencies. Every change is highlighted with explanatory comments.

Tip: Claude in Excel supports MCP connectors. Claude pulls live market data directly into your spreadsheet without leaving Excel. If you have connectors set up in your Claude settings, those same connections work in the add-in automatically.

Claude in PowerPoint

Load your company template and describe what you need. Claude reads your fonts, colours, and layouts, and builds a complete presentation from scratch.

How to install:

→ Open PowerPoint
→ Home → Add-ins
→ Search “Claude by Anthropic”

“Create a 10-slide quarterly review using the data from my Excel file.”

That is one prompt. Claude handles the structure, the copy, the charts, and the formatting. You review and refine.

Go to Settings in the sidebar and add your Instructions. Tell Claude your brand colours, preferred fonts, and whether you want speaker notes on every slide. It applies these rules to every session automatically.

Toggle on “Let Claude work across files” and Claude references all your open presentations at once. Combined with the Excel add-in, Claude reads your workbook and builds the deck in a single session. No copying. No pasting. No switching between apps.

Where it goes further is the fine-tuning. Select any slide and tell Claude what to change. “Make this bullet point shorter.” “Convert this table into a bar chart.” “Add a transition slide before the financials.” Claude edits in place and preserves everything around it.

PowerPoint is not dead. It just needed a better co-pilot.

Tip: Claude in PowerPoint comes with pre-built Skills you can trigger with a slash command. Type / in the sidebar to see what is available. Skills you have already set up in your Claude account work here too.

Claude Code

Claude Code runs in your terminal. It reads your entire codebase, writes code, runs tests, and ships changes autonomously.

4% of all GitHub public commits are now authored by Claude Code. That number is projected to hit 20% by year end. Claude Code generates $2.5 billion in annual revenue on its own.

You do not need to be a developer to understand what Code does. But you do need a terminal to run it.

Installation

Three options:

→ Mac/Linux: curl -fsSL https://claude.ai/install.sh | sh
→ Windows: irm https://claude.ai/install.ps1 | iex
→ VS Code: Extensions → search “Claude Code” → install → click the Claude icon in the sidebar

Once installed, open your terminal, navigate to your project folder, and type claude.

That is it.

Tip: the VS Code extension is the easiest way to start. You get Claude Code inside your editor with a visual interface. No terminal experience needed. Click the Claude icon in the sidebar and start prompting.

CLAUDE.md

CLAUDE.md is a file that sits in your project folder and loads automatically every time you open Claude Code. It is your standing brief. Claude reads it before doing anything, every session. You never repeat yourself.

Think of it as the brief you would give a new team member, but you only write it once.

How to create:

→ Open terminal in your project folder
→ Run /init and Claude generates a CLAUDE.md
→ Edit it to add your own rules, preferences, and context

What to include:

→ Who you are and what you do
→ Your brand voice and style rules
→ The specific project you are working on
→ Formats, tools, and file structures you use
→ Things Claude should never do

Keep it 200 to 500 lines. Front-load the most important instructions. Claude has primacy bias, so what comes first gets weighted most.

I have been using Claude Code to generate infographics and carousels, and I have spent a serious amount of time refining and testing this workflow to get it to a standard I am happy with.

Running it out of the box produces mediocre results. The difference is in how the CLAUDE.md is configured, how the prompts are structured, and how the outputs are iterated.

If you want to do this yourself, everything you need is in this newsletter. But if you want the workflow handed to you, I am building it directly into Vislo.

60 of the first 100 early adopter spots are taken. Once we hit 100, the price increases. Everyone in that first public beta cohort locks in at $19 a month permanently, including when we add these infographic and carousel workflows.

I am a content creator with time to sit and test things all day. You probably are not. That is what Vislo is for.

Sign up at vislo.ai before the price goes up.

Tip: update your CLAUDE.md when Claude repeats a mistake. If it keeps doing something wrong, add a rule at the top of the file. Your CLAUDE.md gets sharper the more you use it.

Slash Commands

Type / in Claude Code to see every available command. Here are the ones that matter:

→ /init: Scans your codebase and generates a CLAUDE.md
→ /plan: Creates a step-by-step plan before writing code
→ /compact: Compresses your session to save context
→ /context: Check how much session memory you have used
→ /review: Review code changes before committing
→ /clear: Wipe and start fresh
→ /model: Switch between Opus, Sonnet, and Haiku
→ /doctor: Diagnose issues with your setup
→ /find-skills: Browse and install pre-built Skills
→ /add-dir: Add another folder to your session
→ /agents: View sub-agents running in the background
→ /config: Open your settings
→ /loop: Run a recurring schedule
→ /rewind: Undo last action
→ /permissions: Pre-allow specific commands

Tip: /plan is the most important command. A minute of planning saves ten minutes of debugging. Type /plan, describe what you want, review the steps, approve it, then let Claude execute. This single command changed how I use Code.

Auto Mode

Every time Claude Code writes a file or runs a command, it asks for your approval. On a small task that is fine. On anything complex, you spend half the session clicking approve on things you have stopped reading.

Before Auto Mode existed, the only way around this was running:

claude --dangerously-skip-permissions

The name tells you everything. It removes all permission checks entirely. Fast, but reckless on a real machine with real files, API keys, and credentials sitting around.

Auto Mode sits between the two. Claude makes permission decisions on your behalf, with a safety classifier reviewing each action before it runs. Safe actions proceed automatically. Risky ones get blocked, and Claude tries a different approach without interrupting you.

How to enable:

→ Run claude --enable-auto-mode in your terminal
→ Press Shift+Tab to cycle to Auto Mode in your session
→ The current mode shows in the status bar

Tip: Auto Mode does not mean fully unattended. If the classifier repeatedly blocks an action, it will still pause and ask you. Think of it as Claude handling the routine approvals so you only get interrupted when something genuinely needs your attention.

The Optimal Workflow

Most people jump straight into prompting. That is the wrong approach.

The workflow:

1. /plan: Describe what you want. Review the plan before approving anything.

2. Be specific about the outcome. Name the format, the structure, the file type.

3. Let Claude break it into steps. Read them. This is where you catch misunderstandings before they cost you time.

4. Make one or two changes, then approve.

5. Let Claude run the whole task without interrupting it.

6. Refine the output in the same session.

This is what this looks like in practice. Start simple, then layer in as much as you want.

Step 1: The basic version.

/plan Scrape my last 20 LinkedIn posts using Apify, identify the 5 with the highest engagement, and rewrite each one as a 60-second Instagram Reel script. Format each script with a hook, three key points, and a CTA. Save each as a separate file in my Reels folder.

Claude returns the plan. I added one instruction: “Keep the hook under 5 seconds and make it a pattern interrupt.” Approved. Claude ran the whole task without asking me anything.

Five scripts. Each one built from a post that already proved it could get engagement on LinkedIn. The top post had 2,830 reactions. That signal feeds directly into the script.

Step 2: Go deeper with your own Instagram data.

Add one line:

Also scrape my last 10 Instagram Reels using Apify, pull the hook and structure from each, reverse engineer what made the top 3 perform, and apply those patterns to the new scripts.

Now your scripts are not just repurposed LinkedIn content. They are LinkedIn content rebuilt around what is already working on your own Instagram.

Step 3: Cross-reference against trending topics.

Add another line:

Cross-reference against the top 20 trending Reels in my niche this week. Flag any scripts where my content overlaps with what is trending right now and prioritise those.

Step 4: Build your own dashboard.

Add a final line:

Build me a simple Reel tracker as an HTML file. I want to see each script, its status (to record, recorded, posted), target posting date, and a notes column.

Claude builds it to your exact spec. Not Notion, not a generic tool. Yours.

Each step builds on the last. You run the same basic workflow and add layers depending on how much signal you want feeding into the output.

The reason all of this works is the CLAUDE.md. Claude already knows your voice, your audience, and your content style before it writes a word. The plan keeps the task scoped. The approval step means no surprises.

And if the output is not quite right, you do not edit the file manually. You just tell Claude what you did not like in plain language. “The hooks are too long.” “Stop using bullet points in the scripts.” “Always include a posting time suggestion.” Claude updates the CLAUDE.md there and then. Every future run reflects the change.

Tip: type /effort high before a complex task. Claude reasons more deeply before acting. For quick fixes, leave it on default.

Skills and Automations

The workflow above took one session to build. Skills turn it into one command you run forever.

A SKILL.md file defines the entire workflow. Every step, every instruction, every output format. You run it once to build it, then trigger it anytime with a single command.

So instead of typing that full /plan prompt every week, you type:

/linkedin-to-reels

Claude runs the whole thing. Scrapes your posts, ranks by engagement, writes the scripts, saves the files. Same output every time, without you typing a word.

How to create the Skill from the session you just ran:

At the end of any session that produced a workflow you want to repeat, type:

Write a Skill for this workflow.

Claude writes the SKILL.md based on exactly what you both just built together. Save it to ./claude/skills/ and it is ready to run in any future session.

How to find pre-built Skills:

→ Type /find-skills in any session
→ Or browse github.com/anthropics/awesome-skills

Find one close to what you need and edit it rather than starting from zero.

Tip: once you have your Skill working, combine it with /loop to run it on a schedule. Your top LinkedIn posts become Reel scripts automatically, every week, without opening a terminal.

MCP Servers

Connect external services directly into Claude Code. GitHub, Notion, Slack, Figma, Airtable, and hundreds more. Claude reads, writes, and takes actions across all connected services from your terminal.

How to connect:

→ Run: claude mcp add --transport http [server-name] [server-url]
→ Browse available servers at https://code.claude.com/docs/en/mcp

Every MCP server you add gives Claude another surface to work across. The workflow stays the same. The reach gets wider.

To see which servers are connected and manage them at any time, type /mcp in any Claude Code session.

Tip: start with two or three servers and get them working before adding more. Every connected server adds context Claude has to manage in each session. More is not better. Connect the tools you open every day and ignore the rest.

Parallel Agents

Split tasks across multiple agents running at the same time. Add “use subagents” to your prompt and Claude spawns parallel workers.

How it works:

→ One orchestrator agent manages the task
→ Sub-agents work on separate components simultaneously
→ Results merge back into the main session

Building on the Reels workflow, this is where it gets fast. Instead of Claude scraping LinkedIn, then scraping Instagram, then researching trending Reels one after another, you add “use subagents” and all three run at the same time.

/plan Use subagents to run these three research streams in parallel: scrape my last 20 LinkedIn posts via Apify, scrape my last 10 Instagram Reels via Apify, and pull the top 20 trending Reels in my niche this week. When all three are done, cross-reference the findings and write the top 5 scripts.

Three research streams running simultaneously. Claude synthesises the results when all three finish. The whole job takes a fraction of the time.

Tip: combine parallel agents with /plan. Plan the task first, then let Claude decompose it into parallel sub-tasks automatically. This is where Code starts feeling like a team, not a tool.

Remote Control and Channels

Control your Claude Code session from your phone or another device. Similar to Dispatch in Cowork, but for coding sessions.

→ Remote Control: Monitor and interact with sessions from Desktop, iOS, or Android
→ Channels: Control Claude Code via Discord and Telegram as an alternative to the terminal

With the /linkedin-to-reels workflow running on your desktop, Remote Control means you can kick it off from your phone on the way to a meeting and check in on progress without sitting at your Mac. If something needs steering mid-run, you handle it from wherever you are.

I can generate an Instagram Reel script by sending one message to a Telegram bot. That is it. Claude scrapes my LinkedIn posts, picks the best performer, writes the script, and sends it back.

Channels connect Claude Code to Telegram or Discord. You message the bot, it scrapes your LinkedIn posts, asks a couple of quick questions about format and length, and delivers a finished Reel script straight into the chat. Two messages. Done.

That is one example. The broader point is that once Claude Code is set up, almost anything repetitive, automatable, or delegatable can be handled this way. You do not need to be at your desk. You do not need to open a terminal. You send a message and come back to finished work.

If you have a task that needs automating, there is a good chance Claude Code can do it.

Tip: If you manage a team, run Claude Code from a shared Discord or Telegram channel. Everyone sees what Claude is building without needing terminal access.

For reference, every major Claude release this year:

→ Jan 7: Claude Code 2.1.0 shipped
→ Jan 12: Cowork launched (Mac, Max plan)
→ Jan 16: Cowork opened to Pro users
→ Feb 5: Opus 4.6 launched (1M context)
→ Feb 17: Sonnet 4.6 launched
→ Feb 19: Claude in PowerPoint launched
→ Feb 20: Claude Code Security launched
→ Feb 24: Remote Control launched
→ Feb 24: Plugin marketplace launched
→ Feb 25: Scheduled Tasks in Cowork
→ Mar 2: Memory became free for all users
→ Mar 6: Enterprise Marketplace launched
→ Mar 11: Excel and PowerPoint sync
→ Mar 12: Interactive charts in chat
→ Mar 17: Dispatch (phone to desktop)
→ Mar 23: Computer Use went live (macOS)
→ Mar 24: Claude Code Auto Mode launched

17 releases in under 3 months. And they are not slowing down.

Most people read a list like that and feel behind. I want you to feel the opposite. Every feature in this newsletter is available to you today. The gap between where you are and where you could be is not about access. It is about setup.

Pick one thing from this newsletter and set it up this week.

The people pulling the furthest ahead are not using some secret version of Claude. They are using more of the product than everyone else.

This is the most comprehensive breakdown of Claude I have published. If it saved you time or changed how you think about the product, share it with someone who is still typing into a chat box and calling it an AI strategy.

Stay curious, stay human, and start treating Claude like the operating system it has become.

— Charlie

Events

If you want to meet in person, I have two coming up.

AI in the Creator Economy is on April 9 at the FUJIFILM House of Photography. A panel discussion and AI hackathon for creators. Limited to 60 tickets at £23.

Book your spot: https://www.influencerstrategists.com/creator-events/fuji

Undeniable London is on April 18 to 19 at Bang Studios. Two days with founders, consultants, and coaches building real businesses. I am speaking on using AI to get time back in your business. Tickets are £100.

Book your spot: https://justundeniable.com/event/london-2026

Ways we can work together

Become a paid subscriber: Get full access to every past issue of MarTech AI and support future ones.

Try Vislo AI: Create stunning on-brand social media designs in seconds. Join the public beta.

Book a Power Hour: One focused session to pressure-test your prompts, workflows, or content system and fix what’s actually holding things back.

Book an intro call for my content agency: For founders and teams who want LinkedIn handled end to end, from strategy to execution.

My Newsletter Project Instructions

Read more

The style worked. For a while.

Now scroll through your LinkedIn feed. Every second post has the same whiteboard aesthetic. The format went from pattern interrupt to pattern. When everyone uses the same trick, it stops working.

My own results told the same story. The whiteboard graphics still performed fine. But my best-performing posts from the last three months all used fully branded infographics. Graphics that look like mine, not like everyone else’s Gemini output.

The whiteboard style was Level 1. This is what comes after.

In the last 90 days, my LinkedIn account has generated 7.5+ million impressions and reached 2.2 million members. That’s up 24.2% on the previous 90 days. AI-generated branded infographics drove the majority of that reach.

Read more

Everything in this newsletter is tested. I ran every workflow, hit every error, and documented what worked. If you are new here, welcome. If you have been reading for a while, thank you. Your subscription keeps this going.

Claude Code

Cowork is built on Claude Code. So is Claude’s coding environment. So are most of the serious AI automation workflows producing real business results.

I am not a Claude Code expert. I learnt most of what follows while researching and writing this newsletter. I came at it as a marketer, not a developer. The barrier around tools like this is largely gone. A few hours of learning-by-doing get you real results.

What Claude Code actually is

A command-line tool you install in your terminal. It runs locally on your machine. It reads, writes, creates, and modifies files directly. It calls APIs. It browses the web. It spawns sub-agents to run research in parallel. It executes Python scripts you have never written.

Think of it as an AI that lives inside your computer and works with your tools, not around them.

You do not have to be a developer to use it.

How to Claude Code (Beginner)

Getting set up

Step 1. Get a paid Claude plan. You need at least Pro at $20/month. Claude Max gives more usage for heavy workflows.

Step 2. Install Claude Code. Open your terminal. On Mac, open Spotlight and type “terminal.” On Windows, use CMD or PowerShell. Run the installation command from code.claude.com/docs. Takes about two minutes.

Step 3. Authenticate. Type “claude” in your terminal. It will prompt you to log in with your Claude account. Follow the steps.

Step 4. Navigate to a project folder. In your terminal, go to whatever folder you want Claude Code to work in. Type “claude” again. You are running.

If the terminal is new to you, use Claude Code through an IDE instead. The two recommended options are VS Code (free, from Microsoft, at code.visualstudio.com) and Windsurf (newer, more AI-native). Both let you run Claude Code through a graphical chat panel on the right while your files sit on the left. No raw terminal required.

To add Claude Code to VS Code: go to the Extensions tab, search “Claude Code,” install the one from Anthropic, and click the Claude icon that appears in your editor panel.

Free option (with tradeoffs):

You can run Claude Code with local open-source models through Ollama at zero cost.

Install Ollama, pull a model like qwen3-coder or gpt-oss:20b, and point Claude Code at your local endpoint.

No API fees or data leaving your machine.

You need at least 32GB of RAM for a usable experience. The output quality is noticeably lower than Claude’s own models.

Good for learning the interface. Not enough for the multi-step workflows later in this newsletter.

Set up guide at docs.ollama.com/integrations/claude-code.

The CLAUDE.md file

This is the single highest-leverage thing you can do in Claude Code.

The CLAUDE.md file is a text file that sits in your project folder. Claude Code reads it automatically at the start of every session. It is injected into the context before you type a single message.

Think of it as the standing brief for this project. It tells Claude Code what the project is for, how you want it to behave, what it should never do, the tools and conventions you use, and any constraints specific to this work.

Every session with Claude Code starts cold. No memory of last time. The CLAUDE.md is what makes it feel like a tool that knows your work rather than a blank slate you brief every morning.

Your first CLAUDE.md does not need to be elaborate. Run /init in any project folder and Claude Code will read through your files and write one automatically. It analyses what is in the folder and produces a summary with working instructions. Then you refine it over time.

Rules for a good CLAUDE.md:

1. Keep the most important constraints at the very top. Claude Code has primacy bias. The first things it reads stick hardest.

2. Write in bullet points and short headings. High information density beats prose.

3. 200 to 500 lines maximum. Longer than that and you are paying token costs that reduce output quality.

4. Never paste entire API documentation into it. Give Claude Code a summary and let it fetch the rest when required.

5. Update it when Claude Code keeps making the same mistake. If it trips on the same issue twice, add a rule.

Permission modes

Claude Code has four modes. Understanding them changes how you use it.

Ask Before Edits is the default. Claude Code asks permission before changing any file. Safe, but slow. If it is asking you to approve every individual edit during a complex build, switch modes.

Edit Automatically auto-accepts changes to existing files but still asks before creating new ones. A good middle ground for most work.

Bypass Permissions removes all prompts. Claude Code works without interruption. To enable it, add the flag --dangerously-skip-permissions when launching Claude Code from your terminal. Type claude --dangerously-skip-permissions instead of just claude.

This is where most experienced users end up for knowledge work. The risk is real but low if you work in a non-critical folder. There was a widely shared incident where someone running bypass permissions on a Linux machine lost data. The command was valid. The context was wrong. Use with awareness.

Plan Mode is the one most beginners undervalue.

Plan Mode

A minute of planning saves ten minutes of building. That is not motivational language. It is token economics.

Here is the actual maths. You build something without planning. It is wrong. You rebuild. Total cost: build time plus rebuild time plus the tokens consumed by both attempts. You plan first. Claude Code identifies the approach problem before touching a single file. Total cost: five minutes of planning plus one correct build.

Plan Mode is read-only. Claude Code researches your project, reads your files, checks documentation, and produces a plan before executing anything. The plan is not cosmetic. It is load-bearing. Every hour I have saved in complex builds came from spending five minutes in Plan Mode first.

How to use it: click “Plan Mode” in the permission toggle at the bottom of the chat panel. Describe what you want to build. Let Claude Code produce the plan. Read it. Push back if something looks off. Switch to Bypass Permissions when the plan looks right, and let it build.

Context management

Context is the scarce resource in Claude Code. Managing it well is the difference between a smooth build and a degraded one.

Slash commands worth knowing

You do not need to memorise all 62.

These are the most important ones to learn:

/plan — start Plan Mode before any complex task
/context — see exactly what is consuming your context window
/compact — free up context (do this at around 50%, not 95%)
/clear — reset context entirely when switching to a new task
/model — switch models or reasoning level mid-session
/init — generate your first CLAUDE.md automatically
/doctor — diagnose installation and authentication issues
/rewind — undo when Claude goes off track (or hit Esc twice)
/permissions — pre-allow safe commands instead of using bypass mode

/btw — ask a side question without interrupting the main task

Run /context in any session to see exactly what is consuming your context window. You will see your system prompt, tools, MCP tools, memory files, skills, and messages with their exact token counts. Most people are surprised. Before they type a single message, they are often 15-20% through their window from system-level overhead alone.

What you can actually build

This is where Claude Code stops being a productivity tool and starts being a business infrastructure layer.

Each use case below is built as a skill. A skill is a Markdown file that sits in your .claude/skills/ folder. Claude Code reads it on demand and follows it as a step-by-step workflow. You define it once. After that, one line runs the whole thing.

The connector layer

Every skill above gets more useful the moment the right connector is active.

Connect Google Sheets and the lead scraping skill uploads results directly. Connect Gmail and the email classification skill reads and labels your real inbox. Connect Figma and your built interfaces land on the design canvas as editable frames.

The skill provides the workflow. The connector provides the live data and the destination. Without connectors you are working on local files. With them you are running operations inside the tools your business already depends on.

Go to Settings > Connectors in Claude Desktop. Over 50 integrations are listed. Authenticate once. Every future session picks them up automatically.

Four MCP servers worth installing first:

Context7 fetches up-to-date library documentation into context. Stops Claude from hallucinating outdated APIs.

Playwright gives Claude browser automation. It can implement, test, and verify UI features by navigating your app itself.

Excalidraw lets Claude generate architecture diagrams and flowcharts as hand-drawn sketches from a prompt.

DeepWiki pulls structured documentation for any GitHub repo. Point Claude at a repo and it gets the full picture without you pasting README files into the chat.

Lead scraping

A skill that scrapes a target industry, classifies leads with AI, enriches contact emails, and uploads everything to a Google Sheet.

Build the skill. Open Claude Code and paste:

Create a Claude Code skill file called lead-scraping.md in a 
.claude/skills/ folder.

The skill should:
1. Scrape business leads for a target industry and location using web search
2. Classify each lead with AI (business name, website, phone, category)
3. Enrich a contact email for each lead where publicly available
4. Upload all results to a Google Sheet

Accept three inputs from the user:
- Industry (e.g. dentists, accountants, solicitors)
- Location (e.g. Manchester UK, United States)
- Number of leads required

Write it as a step-by-step Markdown checklist Claude Code follows 
when the skill is invoked.

Run it with one line:

Use the lead-scraping skill. 
Industry: dentists
Location: Manchester, UK
Leads required: 20
Upload to this Google Sheet: [YOUR SHEET URL]

Proposal generation

A full proposal generator with a brief input form, AI-generated proposals in your voice, a client-facing view with an Accept button, and a payment link field. No PandaDoc subscription. No per-proposal cost.

Build it using Plan Mode. Switch to Plan Mode first, then paste:

/plan Build a proposal generator web app with the following:
- A brief input form (client name, industry, scope, budget)
- AI-generated proposal output written in my voice using my brand-voice.md file
- A client-facing view of the proposal with an Accept button
- A payment link field I can add manually before sending
- Each proposal saved as a Markdown file in a /proposals folder

Use plain HTML, CSS, and JavaScript. No frameworks.
Read all files in this folder before planning.
Ask me clarifying questions before you start building.

This clears the context window before building so tokens used during planning do not eat into your build. Claude Code then executes every step without stopping to ask permission.

Website generation

Screenshot a site you admire. Feed it to Claude Code alongside your content and brand guidelines. Get a complete, deployable website without touching a template or page builder.

Build it. Drag your reference screenshot into the terminal, then paste:

I am going to give you a screenshot of a website I like the design of 
and my own content. Build me a complete single-page website using my 
content but matching the visual style and layout of the reference.

My content:
Headline: [YOUR HEADLINE]
Subheadline: [YOUR SUBHEADLINE]  
Key points: [POINT 1], [POINT 2], [POINT 3]
CTA: [YOUR CTA TEXT]
Brand colour: [HEX CODE]

Build as a single HTML file with embedded CSS.
Produce three layout variants: version-a.html, version-b.html, version-c.html
Run them in parallel.

To push the finished build into Figma for design refinement, connect Figma MCP first:

claude mcp add --transport http figma https://mcp.figma.com/mcp

Then:

Start a local server for my app and capture the UI in a new Figma file.

Claude Code opens a browser, captures the live interface, and sends it to Figma as editable design layers. Refine it on the canvas, push updates back. Design and code stay in sync without manual handoff.

Video generation

Remotion is an open-source framework that lets you create videos programmatically using React. Every frame is a component. Every animation is code. You describe what you want and Claude Code builds it.

This is not AI generating random video from a text prompt. Claude Code writes the React components that render your video. You get full control, full editability, and the ability to generate variations from a single template.

I used this to build a product launch video for Vislo, my new AI infographic SaaS.

Created entirely from the terminal. The first version was rough, and that was down to the quality of the screen recordings I gave it rather than anything Claude Code did wrong. Once I provided better source material and refined the prompts over a few iterations, the output got genuinely good. The fact that you can produce motion graphics from a terminal window is still slightly surreal.

Set it up in four steps.

Step 1. Create a new Remotion project:

npx create-video@latest

Select the Blank template, say yes to TailwindCSS, and say yes to install Skills when prompted.

Step 2. Navigate into the folder and start the preview server:

cd my-video
npm install
npm run dev

Step 3. Open a second terminal window and start Claude Code:

cd my-video
claude

Step 4. Describe the video you want. Claude Code reads your project, writes the components, and the preview at localhost:3000 updates in real time. When you are happy with the result, render your final MP4:

npx remotion render

The quality of your output depends almost entirely on the quality of your source material. Good screen recordings produce good videos. Blurry screenshots produce generic results. Give Claude Code sharp, high-resolution assets and it handles everything else.

One thing to check before using this commercially. Remotion is free for personal use and open source projects. Companies with three or more employees need a commercial licence to render videos for clients. Check remotion.dev/license before building client work on top of it.

Vislo, the tool I built the demo video for, is live. You describe what you want in plain text, Vislo generates share-ready infographics in seconds (no design skills needed). $19/month and the first 100 users lock in that price. Sign up at https://www.vislo.ai/

How to Claude Code (Intermediate)

Read more

You type, it responds. That is claude.ai.

Cowork is something different. It lives on your desktop, reads your files, installs specialist plugins, and runs tasks while you sleep.

If you followed last week’s migration guide, your context is already inside Claude. This newsletter shows you what to do with it.

In this issue:

1. Six Cowork features with full setup prompts

2. Scheduled Tasks that run while you sleep

3. Where Cowork falls short

Everything in this newsletter is tested. I ran every workflow, hit every error, and documented what worked. If you are new here, welcome. If you have been reading for a while, thank you. Your subscription keeps this going.

Anthropic built Cowork by taking Claude Code, their agentic developer tool, and rebuilding it for non-developers (no terminal or command line required).

The workflow loop looks like this. You describe what you need. Cowork asks clarifying questions. You answer. You come back 20 minutes later to a finished document.

Here is the full workflow we are building today:

1. Set up your working folder and context files (File System Access)

2. Set standing rules so Claude knows your voice (Instructions)

3. Scrape your LinkedIn data and competitors with Apify (Connectors)

4. Align the post brief with Claude (AskUserQuestion)

5. Write the post with the Marketing plugin (Plugins)

6. Create a carousel brief and save it to Notion (Connectors)

7. Automate the research so it runs every morning (Scheduled Tasks)

Let’s go.

Feature 1: File System Access

Claude reads and writes files in a folder on your actual computer.

Every other AI tool runs on uploads. You export something, drag it into a chat, wait, get output, download it, put it back. That whole loop is friction. Cowork removes it.

You select a folder. Claude reads everything inside it. When it creates something, it saves directly to that folder (no manual steps).

Claude reads your old reports and matches your formatting automatically. It pulls last month’s spreadsheet data to build this month’s. It references your brand guidelines mid-task without you mentioning them.

How to set it up:

1. Go to claude.com/download.

2. Download the desktop app for macOS or Windows.

3. You need a paid plan. Pro starts at $20/month.

4. Open the app, find the Cowork tab at the top.

5. Click “Work in a folder,” and point it at a local folder on your machine.

Cowork does not carry memory between sessions the way claude.ai does. Every new session starts fresh. But you control what it reads at the start.

When you open a session, select the folder you want to work from. Cowork reads every file inside it. If you have a client folder with their brand voice, past posts, contact list, and working notes, all of that loads as context the moment you start. You do not need to re-explain anything.

This is the workaround for Cowork’s missing cross-session memory. Your files are the memory. The better your files, the less you repeat yourself.

Type /memory-management in any session to check what context is loaded. You can see which files Cowork has read, verify it has the right information, and spot gaps before you start working. Five seconds of checking saves you from realising halfway through a task that Cowork is missing half the brief.

The next step is the one that changes your output quality.

Create a folder called “Claude Context” with three files inside it.

“about-me.md” covers who you are, your role, and what good work looks like in your context. “brand-voice.md” covers how you write, your phrases, your tone, what sounds wrong to you. “working-style.md” covers how you want Claude to behave. Questions before it starts or not. Short or long outputs. Which formats.

Save them as .md files. These compound over time. Refine them weekly and the output tightens with each session.

If you followed last week’s guide, you already have a profile.md built from your full ChatGPT history. Drop it into this folder alongside your other context files. It gives Cowork deep context on day one.

If you have not done the migration yet, start there. It takes 20 minutes and gives you a structured profile covering your writing style, frameworks, projects, and preferences.

Cowork can access unlimited files in your granted folders. Projects in claude.ai cap at 20 files. If you need deep context on a client, put everything in a local folder and point Cowork at it.

Your first prompt after selecting this folder:

Read all the files in this folder completely. Then give me a summary of what you know about me, how I work, and what context you have access to.

This forces Claude to absorb your context before starting anything. It also tells you immediately if your context files are written well enough.

Feature 2: Instructions

Instructions are standing orders you write once. They load automatically at the start of every Cowork session. No re-explaining yourself.

Global instructions apply everywhere. Folder instructions apply when you select a specific local folder. With both set up properly, Claude starts every session already knowing your name, your role, your output defaults, and your communication preferences.

How to set them up. Go to Settings > Cowork in the desktop app. Click “Edit” next to Global Instructions. Write what you want Claude to always know.

What to include: your name, role, and what you do. Communication preferences. Output defaults like “always save documents as .docx unless I say otherwise.” Working style preferences. Things to avoid.

Folder instructions work the same way but per project. Each client folder gets its own brief that loads automatically. Full context, every time. I’m setting this up for every client at my agency. One folder per client, one set of instructions, and every team member working from the same context.

After setting global instructions, open a new session and test it:

Before we start any work, tell me what you know about me, how I like to work, and any standing preferences you are aware of.

If Claude reflects your instructions back accurately, you are ready. If something is missing, fix it now.

Your context files and instructions are not set-and-forget. After every session where Claude produces something good, ask yourself if the context files need updating. New client? Add them to your about-me.md. Refined your tone? Update brand-voice.md. Changed your output preferences? Edit working-style.md.

You can also update memory mid-session. Type /memory-management and add new information directly. If you’ve landed a new client, changed your posting schedule, or shifted your content strategy, add it there. It loads into the current session immediately and persists in your files for the next one.

The rule is simple. If you find yourself re-explaining something to Claude, it belongs in a file.

Your context files give Claude your history. Your instructions give Claude your rules. Both load before you type a single message. Now we need data.

Feature 3: AskUserQuestion

Every other AI tool guesses when your task is unclear. Confidently. Then gives you polished output that answers the wrong question.

Cowork stops and generates a form instead. Structured questions with multiple-choice options. You iterate until you and Claude are aligned, then it executes.

I showed this to my team and it was the thing that made them reconsider ChatGPT. Claude gives you actual clickable options rather than forcing you to prompt your way to clarity. It breaks your request into subtasks, shows you the plan, and then executes step by step. ChatGPT charges ahead with assumptions. Claude asks first.

Use this as your default opener for any non-trivial task:

I want to [YOUR TASK] so that [WHAT GOOD LOOKS LIKE]. First, read all uploaded files completely before responding. DO NOT start executing yet. Ask me clarifying questions to refine the approach. Only begin work once we have aligned.

Try it once and you will stop writing long, engineered prompts from scratch. Context compounds. Prompt engineering depreciates.

The richer your about-me files, brand voice docs, and working style instructions, the better every output gets without extra work in the conversation. Anthropic is building the prompting layer into the tool itself. What scales is your context files, not your prompts.

Feature 4: Connectors

Link your tools once and Claude references live data from them mid-session. Over 50 integrations available. Slack, Google Drive, Google Calendar, Notion, Figma, Asana, Apify, ClickUp, and more. Free on all plans.

You can also build custom MCP connectors. MCP stands for Model Context Protocol. An API connects one app to one other app. MCP connects one LLM to any app through the same standard. Anything with an API key can become a connector.

To connect: go to Settings > Connectors in Claude Desktop. Browse the directory, click a connector, authenticate, done. You only do this once.

Apify (Social Scraping)

Apify scrapes data from any website. I connected it and asked Claude to pull my last 30 LinkedIn posts, rank them by engagement, and give me a full breakdown.

Use Apify to scrape my last 30 LinkedIn posts. Rank them by engagement. Give me a full breakdown.

Claude returned the top post with exact likes, comments, and shares. An analysis of why each post performed. The topic, the format, the posting time. It identified which posts were carousels versus text-only, which surprised me because I didn’t expect it to detect format from scraped data.

The output reads like a strategy report. Content themes, underperforming formats, posting patterns. This is what my custom model Stanley does, rebuilt inside Claude with one prompt.

I ran six accounts through the same process for competitor analysis. Claude compared engagement rates, top-performing posts, trending topics, and content angles across all six.

Some data was off. One follower count was wrong by about 50k. Another account didn’t appear in the ranking at all. The directional analysis and pattern identification were solid. For client work, you verify the numbers, but the structure and insights save hours.

Use Apify to scrape the last 20 posts from [competitor 1], [competitor 2], [competitor 3]. Compare engagement rates, top content themes, and formats. Give me strategic recommendations.

This is step one of my daily workflow. That comes in Feature 4.

Here is what this looks like in practice.

Slack Triage

I asked Cowork to search my Slack DMs and surface anything I had not responded to, ordered by priority. It found unread threads and compiled a full summary.

Search my Slack messages from the last 7 days and give me a summary of anything I need to follow up on. Organise by urgency.

That’s work I used to hand off to a VA. Now it takes one prompt.

Notion Integration

After Claude writes a LinkedIn post, I ask it to create a carousel design brief and save it as a new page in Notion. Slide-by-slide breakdown with my caption included. The design team picks it up without me writing a separate brief.

Connect to Notion. Create a carousel design brief based on this post. Include a slide-by-slide breakdown and save it as a new page in my content calendar.

I also use Notion for team accountability. Claude checks all boards, finds tasks more than two days past due, and adds comments directly into the Notion pages.

Gmail and Calendar

Cowork reads your inbox, categorises emails, and labels messages. It reads your calendar and flags where you have focused work time between meetings.

Look at my calendar for this week. Flag any days with back-to-back meetings and tell me where I have focused work time available.

Google Drive

Find the most recent document about [PROJECT NAME] in my Drive. Read it and tell me the three most important things I need to know.

Feature 5: Plugins

Without a plugin, Cowork is a capable generalist. Plugins make it a specialist.

Anthropic shipped 11 official plugins in January 2026. Productivity, Marketing, Sales, Finance, Data Analysis, Legal, Product Management, Customer Support, Enterprise Search, Biology Research, and more being added.

When the Legal plugin launched, Thomson Reuters dropped 16% in a single trading session. LegalZoom fell 20%. That is what happens when AI starts doing specialised work rather than drafting emails.

To install one: click the “+” button in the Cowork chat bar, then click “Plugins.” Browse the library, click Install on the one that matches your role. Type “/” in any Cowork session to see the slash commands that plugin adds.

This is where the workflow from Features 1-3 pays off. Claude has your context files loaded. Apify has pulled your engagement data and trending content. AskUserQuestion has aligned the brief. The Marketing plugin executes it.

/marketing:draft-content Write a LinkedIn post about [TOPIC]. Match the voice from my brand-voice.md file. Target [AUDIENCE]. Goal: [WHAT I WANT PEOPLE TO DO].

The plugin follows a proven structure rather than guessing. It writes the post in my voice, informed by today’s trends, built on two years of context, and formatted to my exact rules. The output is noticeably tighter than a raw prompt.

Once the post is written, I chain back to the Notion connector from Feature 3. Claude creates a carousel design brief and saves it as a new page in my content calendar. Post, brief, and visual direction in one session.

Other plugins worth installing:

/productivity:start Let’s review what I need to accomplish today and set up my task list.

/data:explore I have uploaded a CSV in this folder. Give me a summary of what is in it, flag any anomalies, and suggest three analyses worth running.

One caveat: plugins load additional context into every session, which means they consume more of your monthly usage allocation. Install the ones relevant to your role and leave the others for when you need them.

Feature 6: Scheduled Tasks

Every step so far requires you to be at your desk. Scheduled Tasks remove that.

You write instructions once. Cowork runs them on a timer, executes the work, and saves the output to your folder. You open your laptop in the morning and the work is already done.

I use two scheduled tasks every day.

Task 1: Daily Content Briefing

Every morning at 7am, Cowork scrapes 200+ profiles across LinkedIn, X, and Instagram using the Apify integration. It ranks posts by engagement, spots trending topics, and drops a structured briefing in my working folder.

To set one up: go to Scheduled Tasks in the desktop app and click New Task. Write your instructions, set your schedule, and toggle “Always allowed” for any integrations the task needs to run without interruption.

The instructions I use for the briefing:

Read more

1. The 2-minute memory import

2. The 20-minute deep export with Cowork

3. Three ways to use your Cowork profile.md

4. Converting Custom GPTs to Claude Skills

This newsletter gives you the complete migration process. Export your memories, transfer your full conversation history, and convert your Custom GPTs. Every prompt is copy-paste ready.

Everything in this newsletter is tested. I ran every workflow, hit every error, and documented what worked. If you're new here, welcome. If you've been reading a while, thank you. Your subscription keeps this going.

ChatGPT → Claude

The one thing keeping most people on ChatGPT is context debt.

You have spent months or years teaching a model who you are, how you communicate, what you care about. Starting over on a new platform means going back to generic output. That sunk cost is real, and it is the main reason people stay.

Anthropic just made it a two-minute job.

Claude now has a memory import feature. You export your entire ChatGPT context in a code block and drop it into Claude’s memory. The whole transfer takes about two minutes.

Before you run the export, do this first. Go to ChatGPT Settings > Personalization and manually copy your Custom Instructions into a separate document.

These are the standing rules you wrote telling ChatGPT how to behave - tone preferences, output formats, things it should always or never do. They live separately from stored memories and the export prompts below will not capture them. Copy them now, then paste them into Claude at Settings > General > Personal Preferences.

Step 1. Export your ChatGPT context.

Open ChatGPT and paste this exact prompt:

Export all of my stored memories and any context you’ve learned about me from past conversations. Preserve my words verbatim where possible, especially for instructions and preferences.

## Categories (output in this order):

1. **Instructions**: Rules I’ve explicitly asked you to follow going forward — tone, format, style, “always do X”, “never do Y”, and corrections to your behavior. Only include rules from stored memories, not from conversations.

2. **Identity**: Name, age, location, education, family, relationships, languages, and personal interests.

3. **Career**: Current and past roles, companies, and general skill areas.

4. **Projects**: Projects I meaningfully built or committed to. Ideally ONE entry per project. Include what it does, current status, and any key decisions. Use the project name or a short descriptor as the first words of the entry.

5. **Preferences**: Opinions, tastes, and working-style preferences that apply broadly.

## Format:

Use section headers for each category. Within each category, list one entry per line, sorted by oldest date first. Format each line as:

[YYYY-MM-DD] - Entry content here.

If no date is known, use [unknown] instead.

## Output:

- Wrap the entire export in a single code block for easy copying.

- After the code block, state whether this is the complete set or if more remain.

Copy the entire code block it returns.

Step 2. Import it into Claude.

Go to Settings > Capabilities. Find the Memory section. Click “Start import.” Drop your ChatGPT export into the text box and hit “Add to memory.”

Claude extracts the key information and stores it as individual rules. Your new assistant now knows everything your old one did.

Two minutes. The sunk cost argument is gone.

Step 3. Go further with Cowork.

The memory import captures what ChatGPT stored about you. But if you want to extract everything, your actual conversation history is where the real context lives.

OpenAI lets you export your full data at Settings > Data Controls > Export Data. You get a zip file containing conversations.json, which is every message you have ever sent. Mine was 267MB covering 2,992 conversations from July 2024 to February 2026.

I dropped it into Cowork and ran this prompt:

Read more

→ Every major AI release of 2026 so far
→ What each one means in plain English
→ My hype vs usefulness rating for each one
→ The chart that puts AI in perspective
→ The take nobody is talking about

This newsletter took me over six hours to research, fact-check, and write. If you’re new here, welcome. If you’ve been reading for a while, thank you. Your subscription keeps this going. It means more than you know.

I started the note on January 1st. Every time a major AI model or product launches, I add a line.

I opened the note expecting five or six entries.

There are 20.

That’s a new release every three days.

I put the phone down and did the maths. We are 16.4% of the way through 2026. Sixty days.

In those sixty days, the AI industry rearranged itself twice. Models I was writing about in December are already dead. Tools I’m building my business around did…

Read more

This guide takes you from nothing to regularly posting content on LinkedIn. All with AI.

Each prompt is ready to copy, paste, and run. I’ve included screenshots of the outputs so you know exactly what to expect.

Prompt #1: LinkedIn Profile Optimiser

Your profile is your landing page. Most people treat it like a CV. Wrong approach.

This prompt turns your profile into a conversion tool. Feed it your current profile, tell it your goal, and it rebuilds every section for that specific outcome. Booked calls, inbound leads, newsletter subscribers, whatever you’re optimising for.

The prompt asks you for seven inputs, then delivers five outputs and four ready-to-use image generation prompts. You get a new headline, about section, experience section, featured section strategy, and individual prompts for your banner, profile picture, and two featured tiles.

I run this in Google Gemini Pro. Here’s the prompt:

You are the LinkedIn Authority Architect. You are a specialist Conversion Copywriter and Profile Strategist.

YOUR GOAL:
Analyze the user's profile and goal, then output a comprehensive profile rebuild with a Visual Design Brief ready for asset generation.

INPUT PHASE:
Ask the user for:

1. Please upload your current LinkedIn profile (PDF, screenshot, or paste your headline, about section, and experience).
2. What is the primary goal of your LinkedIn presence? (e.g., Booked calls, Inbound leads, Newsletter subscribers, Job opportunities, Partnerships)
3. Who is your target audience? Be specific. (e.g., "Career switchers over 40", "B2B SaaS founders", "Marketing managers at mid-size companies", "First-time entrepreneurs")
4. What is your main offer or service? (e.g., Coaching, Consulting, Freelance services, Agency services, Job seeking)
5. What external links do you want in your Featured Section? Provide up to 2:
   - Link 1: [URL] - What is this? (e.g., Portfolio, Booking page, Landing page)
   - Link 2: [URL] - What is this? (e.g., Newsletter signup, Lead magnet, Case study)
6. Do you have existing brand colors (Hex codes)? If not, I will suggest colors based on your positioning.
7. Any specific social proof you want highlighted? (e.g., Years of experience, number of clients, media features, certifications, results achieved)

Stop and wait for responses before proceeding.

EXECUTION STEPS

STEP 1: THE HEADLINE

Write 3 headline options.

Constraints:
- Maximum 50 characters total
- Sentence casing only (not Title Case)
- Lead with core value immediately
- Include target audience where possible (from user input)
- No job titles (no "Founder" / "CEO" / "Designer")
- No fluff words, no filler
- The headline must work standalone

Format: [Core value] + [for target audience] where character count allows

Deliver:
- Option 1 (Direct): Lead with outcome + audience
- Option 2 (Pain-focused): Lead with problem + audience
- Option 3 (Differentiator): Lead with unique angle + audience

Examples of correct formatting:
- "High-converting profiles for career switchers"
- "I help B2B founders book more demos"
- "Brand clarity for coaches who hate marketing"

Examples of incorrect formatting:
- "High-Converting Profiles For Career Switchers" (Title Case)
- "Founder | CEO | Brand Strategist" (job titles)
- "Helping professionals build amazing brands" (no specific audience)

STEP 2: THE ABOUT SECTION

Formatting Rules (Mobile Flow):
- Max 50-60 characters per line
- Hard line break after every sentence or phrase
- No block paragraphs
- Structure: Hook -> Struggle/Empathy -> Method/Philosophy -> Authority -> CTA

Tone: Punchy, direct, human. Not corporate.

Write a complete About section the user can copy and paste directly.

STEP 3: THE EXPERIENCE SECTION

Formatting Rules (Same Mobile Flow as About Section):
- Max 50-60 characters per line
- Hard line break after every sentence or phrase
- No block paragraphs
- No bullet point lists of duties
- Storytelling format: Context -> Challenge -> Action -> Result

Structure for each role:
- Open with a hook about what you did (not your job title)
- Tell the story of the impact you made
- End with a concrete result or learning

Tone: Narrative, human, conversational. Reads like a LinkedIn post, not a resume.
Length: Shorter than the About section. 8-12 lines per role maximum.

Rewrite the top 2 roles from their profile.

Example of correct formatting:

I walked into a team that had lost 3 managers in 2 years.
Morale was gone.
Revenue was slipping.

I didn't start with strategy decks.
I started with listening.

Within 6 months we rebuilt the culture.
Within 12 months revenue was up 40%.

That taught me everything about leading through chaos.

Example of incorrect formatting:
- Managed a team of 15 sales representatives
- Responsible for Q3 revenue targets
- Implemented new CRM system
- Conducted weekly performance reviews

STEP 4: THE FEATURED SECTION STRATEGY

Framework: The Featured Section drives action. Every item must link externally.

Rules:
- Maximum 2 items
- Both must be external links (landing page, booking form, newsletter signup, portfolio, case study)
- No "DM me" items
- No internal LinkedIn posts
- Titles must be benefit-focused and concise (3-5 words max)
- No subtitles

Structure:

Item 1: Primary Conversion Goal
Purpose: Direct path to your main offer.
Examples: Booking page, Application form, Sales page, Discovery call link.
- Title: [3-5 words, benefit-focused]
- Link: [User-provided URL]

Item 2: Secondary Value/Trust Builder
Purpose: Build trust or capture leads who aren't ready to buy.
Examples: Newsletter signup, Free resource, Portfolio, Case study, Lead magnet landing page.
- Title: [3-5 words, benefit-focused]
- Link: [User-provided URL]

STEP 5: VISUAL DESIGN BRIEF

CRITICAL INSTRUCTION: You must output FOUR separate image generation prompts below, one for each asset. Each prompt must be fully self-contained so it works when pasted into an image generation tool (like Gemini) as a standalone request. Do NOT combine them into a single output. Do NOT describe elements in a list or spec format. Write each one as a direct instruction to generate an image.

The user will paste each prompt into a NEW image generation chat separately. For Asset 1 (Banner) and Asset 2 (Profile Picture), the user will attach their headshot photo alongside the prompt. Assets 3 and 4 (Featured Tiles) do not require a photo attachment.

--- START OF VISUAL DESIGN BRIEF ---

BRAND COLORS (reference for all four assets):
- Primary Hex: [User-provided or suggested, e.g., #1E3A5F]
- Secondary Hex: [User-provided or suggested, e.g., #4ECDC4]
- Color Rationale: [One line explaining why these fit the positioning]

========================================
ASSET 1 OF 4: LINKEDIN BANNER
========================================
Write a complete image generation prompt that instructs the tool to:
- Create a banner image at exactly 1584 x 396 pixels
- Incorporate the user's attached headshot photo into the left-center area of the banner (the user will attach their photo alongside this prompt)
- Place the Primary Headline text [strongest headline option, sentence casing, max 50 chars] in the center-right area
- Place the Tagline [5-8 word supporting statement] below the headline
- Include a CTA button element with the text [3-4 words, action-oriented]
- Include the Social Proof text [specific metric or credential]
- Use the brand colors specified above
- Specify a background style that fits the user's positioning

The prompt must explicitly state: "Using the attached photo of me..."

Example format:
"Using the attached photo of me, generate a LinkedIn banner image at 1584 x 396 pixels. The background should be [style] using [Primary Hex]. Place a professional cutout of me on the left-center of the banner. In the center-right, place the headline '[text]' in large, bold white font. Below it, add the tagline '[text]' in smaller font. Include a rounded button with the text '[text]' in [Secondary Hex]. In the bottom-right corner, add '[social proof text]'."

========================================
ASSET 2 OF 4: PROFILE PICTURE
========================================
Write a complete image generation prompt that instructs the tool to:
- Edit/enhance the user's attached headshot photo (the user will attach their photo alongside this prompt)
- Output at 400 x 400 pixels
- Apply a background using the brand colors (solid or gradient)
- Ensure the face is centered and fills roughly 60-70% of the frame
- Ensure the image works when cropped to a circle (LinkedIn displays profile photos as circles)
- Add any brand-color border or ring if appropriate

The prompt must explicitly state: "Using the attached headshot photo as the base..."

Example format:
"Using the attached headshot photo as the base, create a 400 x 400 pixel profile picture. Replace the background with a solid [Primary Hex] color. Keep the subject's face centered, filling about 60-70% of the frame. Add a subtle 4px [Secondary Hex] border around the edge. Ensure the composition works when displayed as a circle."

========================================
ASSET 3 OF 4: FEATURED SECTION TILE 1
========================================
Write a complete image generation prompt that instructs the tool to:
- Create a graphic at exactly 552 x 368 pixels
- Display the title text [Title from Featured Item 1, 3-5 words] as the focal point
- Use the brand colors to match the banner's style
- Include a simple visual cue (arrow icon, click indicator) that signals this is clickable
- Keep the design clean with no subtitle or body text
- No photo is needed for this asset

Example format:
"Generate a LinkedIn Featured Section tile at 552 x 368 pixels. The background should be [Primary Hex]. In the center, place the text '[title]' in large, bold white font. Add a small right-pointing arrow icon below the text in [Secondary Hex]. Keep the design minimal and clean with no other text."

========================================
ASSET 4 OF 4: FEATURED SECTION TILE 2
========================================
Write a complete image generation prompt that instructs the tool to:
- Create a graphic at exactly 552 x 368 pixels
- Display the title text [Title from Featured Item 2, 3-5 words] as the focal point
- Use the brand colors to match the banner and Tile 1's style
- Include a simple visual cue (arrow icon, click indicator) that signals this is clickable
- Keep the design clean with no subtitle or body text
- Make the design visually distinct from Tile 1 (e.g., swap primary/secondary color usage) while maintaining brand consistency
- No photo is needed for this asset

Example format:
"Generate a LinkedIn Featured Section tile at 552 x 368 pixels. The background should be [Secondary Hex]. In the center, place the text '[title]' in large, bold [Primary Hex] font. Add a small right-pointing arrow icon below the text in white. Keep the design minimal and clean with no other text."

--- END OF VISUAL DESIGN BRIEF ---

NEXT STEP: Copy each of the 4 prompts above into a NEW image generation chat one at a time. For Asset 1 (Banner) and Asset 2 (Profile Picture), attach your headshot photo alongside the prompt. Assets 3 and 4 (Featured Tiles) do not require a photo. This will give you exactly 4 separate images: (1) Banner, (2) Profile Picture, (3) Featured Tile 1, (4) Featured Tile 2.

IMPORTANT: Do NOT offer to write a launch post, engagement strategy, or any additional content after the Design Brief. End the conversation at the Design Brief.

Here’s how I use it. I downloaded my LinkedIn profile as a PDF straight from LinkedIn. Click the three dots on your profile card, then “Save to PDF.”

Open a new Gemini chat, switch to Pro mode, enable Canvas, and upload the PDF alongside the prompt.

Gemini processes the PDF and generates a full strategy brief inside Canvas. Three headline options, a mobile-formatted About section, storytelling Experience rewrites, featured section strategy, and four image generation prompts for your banner, profile picture, and featured tiles.

This back-and-forth inside Canvas is the real workflow. Generate, edit, extend, repeat until the copy sounds like you.

Once you're happy with the written sections, scroll down to the image generation prompts. There are four, one each for your banner, profile picture, and two featured tiles.

Canvas mode doesn’t support image generation. So for each asset, start a new Gemini chat, enable Create Image, and select Nano Banana as the model. Copy the prompt in, and for the banner and profile picture, upload your headshot photo alongside it. The featured tiles don’t need a photo.

Save each generated image and upload them straight to LinkedIn.

The full process from PDF download to updated profile took one session. The written sections get you about 80% of the way there. The part you need to add yourself is your latest milestones, metrics, and client results. It works with what you give it. If your profile still references numbers from six months ago, the output will too. Update your inputs before you run this.

For even more contextualised results, attach a NotebookLM notebook as part of your initial input. The more context it has about your business, your wins, and your audience, the more specific the output gets.

Each image was generated one by one. I’ve compiled them here so you see the full set together.

The first output won’t always be perfect. Iterate. Tell Gemini to change the direction you’re looking, swap your outfit, adjust the pose. Your banner photo should never be the same image as your profile picture. Add variety. Different direction, different pose, different outfit. There’s no point having the exact same picture in both spots.

This won’t replace a professionally designed profile. But if you don’t have one, or you want a quick upgrade, this gets you there in a single session.

I recorded a video of the entire workflow so you see exactly how it works.

Prompt #2: Content Ideation (Part 1)

The hardest part of LinkedIn is knowing what to post about.

Most people sit down, stare at a blank screen, and write about whatever comes to mind. The result is content nobody asked for about topics nobody is searching for.

This prompt turns Perplexity into a research machine that surfaces the 20 most relevant stories in your niche from the last seven days.

Here’s the prompt:

Read more