I’m writing these not because I’ve worked it all out, but because I haven’t. We’re in the middle of something that matters, and there’s value for me in recording what it actually feels like to navigate this in real time – the frustrations, the small wins, the dead ends. These aren’t reflections on the future of AI at; they’re my notes whilst trying to get stuff done and rather than keep them private I decided to share. Possibly for no readers but that wasn’t the point.

I’m capturing these as voice notes – usually between meetings or whilst dealing with something else – and having Claude help me structure them. Seems appropriate for a series about how these tools actually work.

ChatGPT stopped being the default

This month was when ChatGPT stopped being my answer to everything. Not because it deteriorated massively, but because the alternatives became good enough at specific things that the general-purpose approach stopped making sense.

Claude became the obvious choice for anything requiring proper writing. Not content generation – actual writing where tone and structure mattered. ChatGPT remained stronger for business thinking, working through strategy, unpacking commercial questions and it was where I had the bulk of my customer GPTs. Gamma had sorted formatting and presentation. Replit was the clear answer for code.

Each tool had found its place. Which sounds sensible until you consider what it meant in practice: constant copy-pasting between platforms.

I’d start something in Claude, shift it to ChatGPT for the business angle, perhaps move it to Gamma if it needed to be presentable, then back to Claude for polish. The switching costs were substantial. Not just the mechanics of cutting and pasting, but the mental overhead of remembering which tool did what well, maintaining context across platforms, losing threads in the transitions.

The only time I didn’t mind that overhead was when getting two tools to red team each other. Feeding back and asking for criticism seemed to work well. Trying it between three did my head in and so generally Claude and ChatGPT it was.

Copilot remained without a clear purpose

Throughout this, I kept trying to find something Copilot was actually best at. There wasn’t anything. There was always a better tool for whatever I was attempting.

This bothered me more than it should have. Microsoft had clearly invested heavily. It was embedded throughout their stack. And yet nothing. Every time I tried it again, I’d think “Claude would have been better for this” or “ChatGPT would have understood first time.”

Copilot wasn’t awful. It just never became the right choice for anything specific and it felt like it had too much to work with.

The fragmentation problem

November’s lesson: having the right tool for each task only matters if the cost of switching doesn’t exceed the benefit.

The costs were real. Time, obviously. But mainly context loss. Every platform switch loses something. The conversation history, yes, but also the implicit understanding that develops when working through a problem with one tool.

I’d already been building custom GPTs to reduce this friction – embedding specific knowledge so I didn’t have to explain the same context repeatedly. But that only worked within ChatGPT. Claude had its own projects, Gemini its own structure, and none of them connected.

The promise had been that AI would make everything easier. I’m not sure that was true – better in many ways but no less difficult or stressful. Perhaps faster and doing tasks I can’t actually do myself.

Why document this

I think we’re living through a substantial shift in how work happens, and the real-time experience will be more valuable than any retrospective analysis. In five years, we’ll have forgotten what it actually felt like to be in the middle of this.

The tools are force multipliers. That’s clear. They make individuals more capable, teams more effective, good ideas more achievable. But they’re not replacements. If you’re viewing them primarily through cost reduction – how many people can we cut now we’ve got AI – then you’re likely getting this wrong.

We have moved past the “one tool for everything” phase into something more complex. Better in some ways – you could get excellent results for specific tasks. Worse in others – managing multiple tools was becoming its own task.

If only there was an answer to that fragmentation …….