# I Changed Tools in a Day. My Work Came With Me.

By Wes Pomeroy. Published July 2, 2026. https://wespomeroy.com/writing/switching-ai-tools/

I'm a Claude person. With Claude it was something like falling in love, fast. Gemini has never been my type. Even for basic chatbot work I've found it underwhelming, and the times I tried something more ambitious with it over chat, it wasn't viable. I use it for images, where the first draft is usually decent, but the moment I start revising it falls apart. Maybe that's on me. Maybe I never gave it enough time or learned the techniques. Nothing about using it ever made me want to.

So when I got to try Google Antigravity at work yesterday - Google's new coding tool, the one built around AI agents doing the work, running on Gemini - I figured I knew how this would go.

It didn't go that way. I've had it about a day, and in that day I've been doing my actual work in it, at close to the quality I get from my usual setup.

## Feature by Feature, It's a Wash

For the things I lean on day to day - having it drive a browser, run commands, read and write my files, keep my projects and their context organized - Antigravity holds up right next to Claude. They trade off. Antigravity feels quicker at getting things done and moving around the machine, while Claude is slower but thinks harder when a problem needs it.

So the first thing is simple: Antigravity is good. A capable second option, and that's good on its own, because more competition means more choice, and it keeps some pressure on price and quality.

## My Setup Came With Me

What I appreciated more, though, wasn't a feature at all. My team and I have built our own way of working on top of Claude - how we manage context, how we keep memory across sessions, how we hand work back and forth. I pointed Antigravity at the whole thing, gave it some context, and it understood how we work and moved it over with barely any friction. It made a couple of mistakes. No more than Claude would have, maybe fewer.

Which means the work we put into that wasn't throwaway. The patterns we'd worked out for getting this kind of AI to actually do the job held up on a completely different tool.

## The Ground Keeps Moving

The tools are going to keep changing under us. The genuinely powerful capabilities - agents that do heavy work with less hand-holding, systems a normal person can pick up without a developer standing next to them - aren't there yet for everyday users. Claude Cowork gets closest to putting this in front of non-technical people, and even it isn't something I could hand my dad and walk away from.

But we can already do a lot right now, even in this hand-built stage. And building things by hand is how the patterns and the capabilities you actually need become clear. You find out what good looks like by making it. So as the platforms catch up, we'll already know what to ask them for.

## Don't Get Attached to the Tool

Don't get attached to a tool, or to a particular way of doing a thing. Get attached to the capabilities you need to do your work.

That attachment is what gives you a say when the tool gets picked. When the business side shows up with no clear requirements, no advocate, no sponsor, the choice gets made on other grounds - what IT wants to support, what Finance wants to pay for. Those are fair considerations. They just shouldn't be the only ones, and in a vacuum they become the only ones. Knowing what you need, and being able to say it plainly, is how you get a say instead of having the decision handed to you.

That's why digging in right now helps. Try the tools. Build things, even the experiments you throw away. You learn what your work actually requires by doing it, not by reading a roadmap, and you learn how to describe it. The next time someone's deciding what the team runs on, you can walk in knowing what the work needs.

The tools will keep coming and going. What it takes to get your work done is steadier than any one of them. Start there. And while you're at it, go try Antigravity. It doesn't suck.