Hey friends,

A couple months ago, I made a call that I had been putting off for a while. I moved our whole organisation off ChatGPT and onto Claude.

Not just me. Not just the AI team. Everyone.

It is not the kind of decision you take lightly. People had built habits, prompts, workflows. ChatGPT had been the default for two years. And yet, sitting here a few weeks in, I can honestly say it has been one of the most useful shifts we have made as a business.

So I want to walk through why, because I think the reasons matter more than the move itself.

Let me get this out of the way first. We did not switch because of the underlying model.

Claude's Opus is a beautiful model. In a lot of writing tasks it outperforms what I get out of GPT. But honestly, models come and go. OpenAI will release something better next month. Then Claude will respond. Then somebody else will leapfrog them both. Chasing the best model is a losing game.

The reason we moved is the layer around the model. The orchestration. The tools. The way the whole system is built to actually do work, not just answer questions.

That distinction is the entire conversation.

This is the clearest way I can frame it.

ChatGPT is a brilliant tool if you want to chat with AI. Recipes, homework help, drafting an email, settling an argument with your partner about who said what in 2014. It is genuinely good for that.

But when you start asking AI to do real work inside a business, to actually complete a task end to end across files, systems, and people, ChatGPT has not been built for that level of operation. Claude has.

That focus shows up in everything they release. Every feature feels designed for someone trying to get work done, not someone trying to get an answer.

The first feature that genuinely changed how we work is something Claude calls Skills.

Here is the simplest way to think about it. A large language model is like a brilliant graduate who just walked into your office. Smart. Capable. But they do not yet know how your business does things. They do not know how you write a proposal. They do not know how you do an SEO audit. They do not know your tone of voice.

Skills are the way you teach them.

You walk through how you do a particular piece of work, get the AI to do it your way once, and then you save that as a skill. From that point on, anyone in the team can call that skill with a slash command and get it done the same way every time.

We have built one called Humanizer. Our content team uses it to take any draft and check whether it reads like an AI wrote it, then strip out the tells. That used to be a back and forth conversation every time. Now it is one keystroke.

The thing that makes this powerful is that you do not need a developer to build them. You describe the work, the AI builds the skill, you save it. That is it.

AI without context is annoying.

You sit there copy-pasting emails into a prompt, dragging files in, screenshotting your CRM, explaining background that the AI should already know. After a while you stop bothering and the tool stops getting used.

Claude's connectors fix this. You can plug it into Gmail, Drive, your CRM, your accounting tool, your design tools, your project management. And for anything that is not on the standard list, there is something called MCP, Model Context Protocol, which lets you connect your own systems.

Now combine that with skills. The SEO audit example I keep coming back to. You build a skill that says, when I ask for an audit, go to Google Drive to look at the format of previous audits, go to the website itself to scan the content, go to Ahrefs through the connector to do the keyword research, then check the CRM for context on this customer's industry, then write the audit in our usual format.

That is not a chat. That is a workflow. And it runs on its own.

Both ChatGPT and Claude have projects, but the experience inside Claude is in a different league.

You create a project, attach the relevant files, connect the relevant Drive folders, write the project instructions, and Claude remembers everything. You can have ten chats running, each focused on a different part of the work, and the context never gets lost.

On the team plan, multiple people can work inside the same project, in their own chats, with shared memory. So Mansi, Tanisha, and I can all be working in the same client project at the same time and we are not stepping on each other or repeating context.

That changes how a team uses AI. It stops being a personal tool. It starts being a shared workspace.

This deserves its own conversation, but briefly.

Claude has a design feature now that lets you load your brand system, your style guide, your colours, your fonts, and produce on-brand designs every time. Wireframes, social posts, brochures, internal docs.

The thing I like most is that what it produces is editable. So you can tweak a heading, change a colour, adjust a width without re-prompting and praying the AI does not give you a completely different image. And if you want to finish in Canva, you can export straight into it.

Our marketing and UX teams have used this more than I expected.

This is the feature that actually changed my mind.

Claude has a desktop app, and inside it there is a feature called Cowork. Cowork lets you point Claude at a folder on your computer and tell it to do something with the files. Then it just goes.

Not "help me with this." But "handle this."

I tested it on bank reconciliation. We have five or six accounts across the family. I dropped all the statements into a folder, gave Cowork access, and asked it to figure out where the money was going, where it was leaking, what we were paying for that we did not need, and to build a forecast.

It went through every statement. It found subscriptions we had forgotten about. It flagged double-ups. It found bank fees we did not know we were paying. It built a twelve-month forecast we had not even asked for. And then it built a dashboard so I could see the whole picture in one place.

That is not a chat. That is a job.

A small tip that has made a big difference for us.

When you are getting Claude to produce a document, ask it to create the document in print-friendly HTML. Not Word. Not PDF. HTML.

The reason is that Claude can do things in HTML that look like a senior graphic designer spent a week on the layout. Properly designed reports, with infographics, charts, structure, and brand consistency. The kind of report a big consultancy would charge you five or six thousand dollars to produce. You can then save it as a PDF and you have a document that is genuinely impressive.

We have used this for proposals, internal reports, client deliverables. The quality difference compared to Word output is significant.

Claude has had artifacts for a while. Small apps, simple dashboards, quick visualisations. But Live Artifacts are something else.

A live artifact is a working dashboard that updates from real data inside your connected systems. So you can ask Claude to build you a profit and loss visualisation, point it at your accounting connector, and now you have a dashboard that refreshes itself.

And the part most people miss, it is not just structured data. It can read unstructured data too. I connected one of my dashboards to our team's Cliq channel, which is similar to Slack. It now reads the messages, understands the context, and surfaces things that would otherwise just get lost in a busy thread.

For a small or mid-sized business, this removes a meaningful amount of the case for separate BI tooling.

I will not go deep on Claude Code in this newsletter, it deserves its own. But the short version is this.

Claude Code is AI that lives inside your computer's terminal. Anything you can do on your computer, Claude Code can do. Read, write, organise, automate, build, deploy. The possibilities are honestly endless once you get comfortable with it.

This is for the more technically curious end of the audience, but I want to flag it because it is part of the reason the move made sense. Claude is not one tool. It is a stack.

I do not want to make this sound like everything is better and there is no friction.

You are giving an AI access to your files, your inbox, your CRM, sometimes your bank statements. That requires real thought about permissions, security, what you connect, what you do not. We had to think hard about who in the team gets access to what.

There is also a learning curve. Cowork is not the same as chat. Skills take effort to set up well. Connectors need to be configured properly. The teams that get the most out of this are the ones who treat it like onboarding a new system, not like installing an app.

If you give Cowork a vague instruction, run it once, and decide it does not work, you have not really tried it.

If I had to summarise the move into one line, it is this.

ChatGPT helps you think faster. Claude helps you get work done without you.

That is a different category of tool. And once you have experienced the difference, going back feels like going from a power tool to a hand tool.

The conversation around AI keeps getting stuck on which model is best. That is the wrong conversation.

The model is becoming a commodity. The orchestration is the moat. The skills, the connectors, the projects, the cowork loops, the dashboards, the design layer. That is where the leverage actually sits, and that is where the gap between a casual AI user and a serious one is going to widen.

Whether you stay on ChatGPT, move to Claude, or use something else entirely is less important than this. Stop treating AI like a smarter Google. Start treating it like an operational layer that can run alongside your team if you set it up properly.

Most people are not going to do that work. The ones who do are going to look like they have an unfair advantage. They will not. They will just have actually treated AI like the tool it has become.

See you next week,

— Aamir

🎧 Listen to the Podcast Episode 1 on: Spotify | Apple Podcasts | YouTube

📘 Book: The CEO Who Mocked AI (Until It Made Him Millions) by Aamir Qutub

🚀 Dumb Monkey AI Academy

📱 Dumb Monkey AI Academy App: Apple | Android