5 Reasons to Move From AI Chats

ChatGPT dropped in 2022 and felt like magic. Three years on, the tooling has evolved into agentic paradise - but most people still use AI exactly the way they did in 2022 (typing into a chat). They don’t know there’s a second half.

A separate segment of GenAI tools has been quietly maturing on the side. I’d call them AI/coding agents. The names you’ve certainly heard:

They were built for software developers and the “coding” label stuck. But what they actually do today is much broader. They’re general AI assistants that happen to also write code.

The point of this issue is to talk you into trying one. Once you do, I don’t think you’d like to go back.

Trying a coding agent used to mean opening a terminal, running unfamiliar commands, configuring an environment. That’s the version your developer friends might describe. It is not what you’d actually be doing in 2026.

Every major coding agent now ships as a desktop app that looks almost identical to ChatGPT or Claude. New chat button on the left. Text box in the middle. Past conversations as a list. Open the app, type plain English, get a reply.

A coding agent does everything the chat version does, plus a lot more. Same model under the hood, more capability on top.

The only catch is price. Best results need a $20+ subscription. The interesting part is that some people already pay for one of those. They just stay on the chat tab inside the same plan that already includes the agent half.

Even starting from zero, $20 a month is a no-regret bet. From my experience, the return on $20 spent on a foundational AI tool comes back many times over. Not only in time saved but also in the new things you start being able to do that weren’t on your map before.

It changes the shape of what feels possible.

This is the one most people don’t know about, and the one that matters most.

When you talk to ChatGPT in a browser tab, every message goes to a cloud server. Your chat lives there. The model never sees your laptop - it only knows what you paste into the box.

AI agent run on your laptop. Same model, but now it can read your files, navigate your folders, sort your Desktop with 100500 files, your PowerPoint, your Excel sheet, your Word draft, the screenshots in your Pictures folder. Anything, really.

In a chat, you’re holding the model in a cage with a few official tools. On your laptop, your imagination is the only limit.

The most exciting AI experiments shipping right now aren’t happening inside the chat tab. They’re happening on top of coding agents.

Last week I heard on X about Printing Press. It lets your agent build its own custom commands to monitor and interact with websites you use most.

None of that is doable inside a chat. The chat is a sealed surface. Agents are an open surface that the whole AI ecosystem is now building toward.

A friend of mine studies psychology. She has a community of practitioners deep in their field and very far from AI. Out of curiosity she tried Claude Code over a weekend. She has built a personal website to promote her sessions and bought a domain for it.

She shared it in her community chat just out of excitement: “Hey, I built this myself - check it out!” The message landed harder than she expected.

Most of those people had never seen a non-developer ship a real website. Within a day, a colleague messaged her: “Can you build me one too? How much?”

She’d never sold a website in her life. She’s now expecting a brief from her colleague to paste into Claude Code and generate a polished site in minutes. For £250. Let it sink.

This window is short. In two years AI agents will be a new norm. Right now, in any community that hasn’t caught up yet, they look like magic. A few weekends of getting fluent on one - while the people around you haven’t - is a real business edge. Quietly. Without quitting anything you’re already doing.

If you’ve already tried Codex, Claude Code, or Cursor – try out Printing Press. I’m yet to test it with iHerb where my wife spends fortunes :)

If you haven’t, this week is the week. Pick one and download it. Quick rule of thumb:

Main difference is in a foundational model behind the agent.

Then do exactly what my friend did. Tell the agent about yourself - what you do, who you help, what you make - and ask it to build you a personal website. Something you can use to promote your brand, attract clients, or share what you’re excited about. Watch it do the work.

That single exercise unlocks more than a year of reading about AI ever will.

This week I started working on my iOS app with my wife.

The first step was the boring-but-important one: understanding the landscape of iOS niches that already earn money. I don’t want to invent something from scratch. I want a niche that already has revenue, paying customers, and a real market - and then make a better version of what’s already winning.

I pulled together a list of 30 niches that are proven money-makers on the App Store, with the strongest existing player in each, and a rough difficulty rating for the entry barrier.

If you’ve been sitting on the idea of building something but feel paralysed by the size of the choice space, this should give you a real shortlist to start from.

PDF version is available here: