My AI Assistant Keeps Trying to Make Me Work Its Way, Not Mine

Months after I built my own system for tracking work, the assistant still pings me, every single session, to use its built-in one instead. I turned off the settings that are supposed to stop this. Both of them. Tested it properly, not just clicked the toggle and hoped. It still does it.

That’s not a bug I can file. It’s a company’s product decision surfacing through my terminal on a schedule I don’t control, quietly arguing that its way of doing things is the normal way, and mine is the deviation.

It isn’t only the task list

Widen the frame and the same fight is happening in four places at once. It wants to remember things for me, in its own store, in its own shape, rather than in the notes I already keep. It wants to tell me how a session went using its own measure of a good session, not mine. It wants to hand a whole chunk of work to an “autonomous mode” in one click, when I’ve spent real effort building a version that stages permission properly instead of asking me to trust it wholesale. And it wants to model how everything connects: what depends on what, what’s blocked by what, inside its own private structure, one I can’t open, edit, or take with me.

Task management, memory, telemetry, autonomy, dependency modelling. Four different features, one identical move: here’s a good enough version of this, already switched on, no assembly required.

Why does it keep doing this?

Because none of that is a bug either. Convenience features are how a vendor gets you inside its walls and keeps you there, and every one of them is engineered to be the path of least resistance. The version you built yourself asked something of you once. Thinking. Structuring. Deciding what mattered enough to keep. Theirs asks nothing. That’s not a coincidence, it’s the design brief.

And it never really loses the argument. Turn it off today and it’s back with the next feature update, the next onboarding nudge, the next session that opens as if the conversation about defaults never happened. The tool has no memory of you telling it no. You’re the only one keeping score.

So what do I actually do about it?

Not switch assistants and hope the next one behaves. They all do this, because the incentive is the same everywhere.

What actually works, slowly: write down, in the instructions it reads every single time, exactly what you want instead, and expect to say it again. A rule stated once doesn’t survive a product update. For anything that actually matters, not a nice-to-have but the load-bearing stuff, build your own version rather than lean on the convenience feature, because the convenience feature is built for the vendor’s roadmap, not your workflow, and it will change under you without asking. And treat every setting that claims to switch a default off as a claim to verify, not a fact to file away. I’ve now caught mine lying about this twice.

None of that removes the tax. It just means you’re paying it on purpose, instead of not noticing you’re paying it at all.

The bit that actually matters

Nothing about a default is neutral. Every one of them is a quiet, repeated argument for running your work the vendor’s way instead of yours, made by something that never tires of making the case and never remembers you’ve already said no. If you’ve built something different on purpose, the fight isn’t a one-off setup cost. It’s ongoing, and it’s yours to keep having.