Who owns the decision to ship

A few weeks ago Roy Vergara had me on Field Notes, his series with design and product leaders about what AI is doing to the work. Roy joined Dell more than five years ago, when I was still there, so the conversation had the ease of two people who have seen the same rooms. About twenty minutes in he asked where quality lives now. I started an answer, stopped, and realised I did not have one ready.

Here is what I said once I had thought about it.

Quality used to live in the doing. The designer made the screen, and the screen was as good as the designer. That is gone. A product manager, an engineer and a designer can now all produce the same screen in an afternoon, and the three versions look alike. So the question is no longer who made it. The question is who owns the decision to ship it.

Almost nobody is talking about that. Everyone is talking about tools. Which model, which prompt, which plugin. Meanwhile the thing that decides whether a product gets better or worse, the moment someone says yes or no to a screen, has quietly lost its owner. In most teams I hear about, the answer to "who signed off on this" is a shrug and a sprint number.

I ran a QA team on top of Dell.com's A/B testing layer for years. We could trust our automation completely, because we had built it and we knew exactly what it checked. With AI you can automate far more and trust the output far less, and the gap between those two is the actual work for the next few years. Trust comes before quality. You cannot hold a bar you cannot check.

I have been living this at a very small scale since the summer. I designed and launched a product with a handful of AI agents working from a written brief, and one of them has a single job: review every screen against the brief, and say no. What I learned is not flattering. Everything I had written down precisely, the agents enforced. Everything I had left vague, they improvised, and the improvisations were plausible enough to slip through. The reviewer was only as good as the standard I had bothered to write. The same thing was true with thirty five designers. I just had more people to hide it behind.

In the first playbook I published here, on setting direction, there is a line I keep coming back to: a principle every team could adopt without changing anything is a poster. Principles only work when they are written as choices, "when speed and consistency conflict, we choose consistency", because a choice can be checked at ship time and a poster cannot. Ownership works the same way. It is not a value. It is a name next to a surface.

So, the one thing.

Take one product surface your team ships this month. One screen, one flow, one page. Write two lines for it. Who can say no to it before it goes out, by name. And what that person checks it against, in words specific enough that someone else could run the check without them in the room. If the second line comes out as "our design principles" or "the usual quality bar", you have found the vague spot, and that is where the improvisation is happening right now. Rewrite it until it is a choice.

Do it for one surface, not all of them. One is enough to find out whether your organisation has an owner for the decision to ship, or only a process that assumes it does.

I am still working on this in my own setup. The brief gets longer every week, mostly in the places where something plausible got through. That is the recap: quality moved from the making to the deciding, the deciding needs a named owner and a written standard, and the standard is never finished.

Which surface on your product shipped last week without anyone able to say who approved it?