An agent is useful exactly as far as you can check its work. If you can't check it, you didn't get faster — you just got sooner to something you'll agree with without looking.
Five years ago I designed wallet creation. I remember how long it took to put it all together and work out how the thing even works. Now an agent does it from a voice request. And it lays out, separately, the states where a passcode is part of registration → I didn't even know those were required.
What I felt was relief. Not a sting, not anxiety — relief: I had just saved myself a pile of time. And I saw things that I, with all my experience, could simply have forgotten.
But it only works one way. Skills and agents are useful first of all when you are a professional yourself — when you can check what you have been handed. If you have experience in the area, it makes you dramatically faster.
Then came a case where I couldn't check. Loyalty program tiers, limits tied to payments. We didn't know how to distribute the limits across tiers, and I leaned on the agent entirely. It found how a large competitor had done it, brought the whole thing back, and built the product logic and the interface on top of that.
I believed it. Not because I checked — because the result satisfied me. And satisfied everyone. Even if we had drifted from the truth somewhere.
It is the same work as if another designer had brought me that solution. I look at it, I don't know what is right, and it broadly satisfies me — I would have done exactly the same.
A good check doesn't have to end in a hundred percent agreement. Sometimes you just tell yourself: I can live with it.
I say no often. And almost always on new design — not on finishing a task, not on improving something that already exists. The problems are legible, exactly the ones a junior designer has: the agent doesn't see the dynamics between screens, doesn't think systemically, forgets the goals. The same things you write in a performance review for the first three years.
The difference is that you no longer have to spend those three years. You can teach your junior very fast, with a single file. And it doesn't disappear for a week to bring back a design with mistakes — an iteration takes minutes.
So I always push to make, to iterate and to check, rather than sit and guess.
In other words → Do more!