Keeping control when everyone starts using AI
Working this way suits me, and I’ve written before about why the accountability has to stay with the person and not the tool. I’ve also watched it carry a whole project, delivering in two months what a client had set aside a year for. The question I get asked most, though, is the harder one: how do you let a whole organisation work like this without it going wrong?
It’s a fair worry. One experienced person keeping a close eye on AI output is one thing. A hundred engineers of varying experience, all moving quickly, is another. Here is what I think keeps it safe, and none of it is exotic.
Accountability doesn’t move
This is the first rule, and the one everything else rests on. Whoever ships the work owns the work, however it was produced. “The AI did that” cannot become an acceptable answer anywhere in the organisation, because the moment it does, nobody is really responsible for anything. Make it explicit: in how you review, in how you talk about output, in who signs off. People need to feel the same ownership of AI-assisted work as they would of something they typed line by line.
Raise the standards, don’t relax them
There’s a temptation to treat AI output as somehow exempt from the usual rigour, as if a machine writing it means it needs less checking. The opposite is true. Testing, code review, documentation, CI/CD, all of it matters more when code is being produced this quickly, not less. These are not bureaucracy to be trimmed in the name of speed. They are how you keep the output honest at the pace AI lets you work. If anything, hold the line harder than you did before.
Make the good patterns shared
Left to their own devices, every team will solve the same problems in slightly different ways, and the quality will scatter. On the project I mentioned we built a shared set of prompts, plugins, skills and standards that everyone worked from, and improved them as we learned. Do that across the organisation. When someone works out a better way to get reliable output, it should become everyone’s way, not stay stuck in one team. Consistency is what stops “fast” turning into “unmanageable”.
Don’t let understanding leak away
This is the risk I worry about most. When AI writes the bulk of the code, it becomes very easy to end up with systems that work but that nobody on the team actually understands, and it creeps up on you.
Two things guard against it. The first is keeping the things that carry understanding genuinely current: documentation, architecture diagrams, clear onboarding paths. None of that is new, but it matters far more now than it did. The second is people. Engineers early in their careers need to build the judgement to challenge, correct and validate AI output, and they won’t get it by waving everything through. That means deliberate mentoring and review, rather than an assumption they’ll absorb it on their own. My own confidence with this comes from years of doing the work by hand first, and we can’t assume the next generation gets that for free. So we have to build it in.
Have an answer for when it breaks
If your delivery now depends on an AI tool, treat that dependency seriously. What happens if the service goes down mid-sprint, or the price rises to a point that changes the economics of how you work? These are not reasons to avoid the tools, but they are reasons to know where you stand: to keep people able to work without the crutch when they have to, and to avoid building yourself into a corner you can’t afford to stay in. A sensible rollout plans for the tool not being there as carefully as it plans for it being there.
Treat it like a team member
None of this is a new discipline. It’s the engineering discipline we’ve always known, applied to a faster way of working. Treat AI as another member of the team: useful, quick, capable, and held to exactly the same standards as everyone else, with its work reviewed and owned by a person every time. Do that, and you get the speed without the mess. Skip it, and you end up with a great deal of code, produced very quickly, that nobody can stand behind.