I have three AIs and no favorite

Every few weeks someone asks me which AI is the best one. They want a name. My answer usually disappoints them: I use three of them every week, I only pay for one, and I have no favorite. Claude writes my code and plans my work. Gemini does my research and my visuals. ChatGPT handles language, the emails, the posts, the wording that has to sound right. Asking which one is best is like asking which person on your team is best. The question assumes one of them should be doing everything.

I tried making one do everything

I didn't start with three. I started the way most people do, looking for the one tool that could carry the whole job, and I gave each of them a fair chance at things outside their strengths.

Code was the clearest test. Gemini and ChatGPT both start well. The first version comes back fast and looks right. Then the request gets longer, the deliverables pile up, and somewhere around the third round they lose the thread. A requirement we agreed on quietly disappears. A fix lands in the wrong place. I find myself explaining the project again from the top, and that's the moment I know I'm managing the tool instead of using it.

Claude fails in the opposite direction. Give it a large codebase and a long list of deliverables and it holds on to all of it. Ask it for a mockup and it's like asking a developer to draw. Everything is technically correct and nothing is inspired. Every screen looks like a settings page.

None of that is a complaint. It's a staffing note.

This is just a team

I've led product teams for a long time, and nobody on a good team does everything. The engineer who can hold a whole system in her head is rarely the person you want designing your brand. The researcher who finds the one insight that changes the roadmap is rarely the person writing the launch copy. Nobody expects them to be, and nobody calls that a flaw. That's the design of the team.

So my AIs work the way a team works. Research starts in Gemini. What it finds goes into a plan in Claude, and the plan becomes code in Claude too. The words that go out to people get shaped in ChatGPT. When something needs to look good, it goes back to Gemini.

And the work that matters most happens between them. None of these tools knows what the others did. I'm the one carrying the context from one to the next, deciding what each one gets, briefing it, and reviewing what comes back before it moves on. That isn't prompting. That's management.

A team that has never met

Managing this team comes down to two jobs, and they're the same two jobs I've always had with people.

The first is finding out what each member is actually good at. The marketing page won't tell you. The work will. I learn it the way I'd learn a new hire: small tasks first, then bigger ones, watching where the quality holds and where it slips. That's how I found out that Gemini and ChatGPT lose the thread on long builds, and that Claude can't design. Nobody told me. And it keeps changing. Every new model version is a new hire with the same name, so the testing never really ends.

The second job is harder: getting them to work toward the same goal when they have no connection at all. They don't share a chat, a memory or a goal. Each one knows only what I hand it. So I do what a good lead does with a distributed team. I write the goal down once, in a short brief, and every one of them gets the same version of it. I define the deliverable before the work starts, so the research knows what plan it's feeding and the plan knows what code it's becoming. And every hand-off happens in writing: the output of one becomes the documented input of the next, never a summary from memory.

Done well, tools that have never met behave like one team. Done badly, you get three excellent pieces of work that don't fit together.

Right now, the only thing connecting them is me. That works with three. I'm not sure it works with ten.

Bottom lines

"Which AI is best" is the wrong question. Ask which one is best at this job.

Every model has a ceiling outside its strengths. Some lose the thread on long work, some can't make anything look good.

The hand-offs are the real work. The tools don't share context. Whoever carries it between them is managing.

A shared goal has to be written down. Tools that have never met only agree on what you hand each of them in the same words.

So no, I don't have a favorite AI, the same way I never had a favorite person on a team. I have a lineup: Claude for code and planning, Gemini for research and design, ChatGPT for language. That's mine, and I know it's a small team. There are dozens of tools out there now, many built for exactly one job, and I've tried only a few. So I'm asking two things. Which tools are you using, and for which tasks? And which one should I be using that I'm not?