Most teams can’t answer that with a number. They can tell you the tools are in use, that engineers like them, and that code gets written faster. Ask whether the engineering backlog is any shorter or the roadmap ships any sooner, and the room goes quiet.
That silence is worth sitting with, because the tools did work. Autocomplete, agents in the editor, whole functions from a comment: generating code got dramatically cheaper. If the backlog didn’t move, the gain went somewhere. Here’s where to look.
Where the velocity went
It got absorbed by the same queue. An engineer who writes a change in an hour instead of a day still has to wait for the ticket to be prioritized, pulled into a sprint, reviewed, tested, and deployed. Generation is one step in that chain, and it was rarely the longest one. The developer productivity gain lands on that one step, and the chain still moves at the pace of the slowest.
Review got heavier, and review didn’t get faster. More code, produced faster, lands on the same reviewers. A senior engineer reading a generated pull request has to check that it reused the right component, matched the conventions, and didn’t quietly break something two folders away. None of that got cheaper, and it all lands on the most senior people in the room. Speed up the juniors and you’ve moved the bottleneck up a level.
Small changes still pay the full process tax. A copy fix, a color change, a new field on a form. The generation step for these was never the cost. The cost is the ticket, the triage, the context switch, the review, and the deploy. Those are fixed overhead, and they’re the same whether the code took five minutes or five hours to write. The backlog is full of this kind of work, and the tools did nothing for it.
Demand expanded to fill the capacity. Faster engineers get asked for more. Product sees the velocity in the editor and adds to the roadmap. This is reasonable behavior from everyone involved, and it means the queue never gets shorter even when throughput goes up.
The audit
If you want to know whether the tools changed anything for the business, three questions will tell you. All three can be answered from your ticketing system and your repo in an afternoon.
- What’s the cycle time from spec to production for a small change? Pick ten changes from the last quarter that a product manager would describe in one sentence. Measure from the day the ticket was written to the day it was live. If that number hasn’t moved in 18 months, the AI coding tools made writing faster and shipping stayed the same.
- Who authored the last fifty merged PRs? If every one came from an engineer, then every change still routes through the same team, no matter how it was written. The tools made that team faster at their own work. They didn’t change who can do the work.
- How much of the backlog is work that didn’t need an engineer to write it? Go through the open tickets and tag the ones that are UI, copy, config, or layout. The ones where the spec is the hard part and the code is the easy part. That share is the velocity you bought and never received.
What I expect you’ll find
I’ll make a prediction, and you can check it against your own numbers. Cycle time for small changes is flat. Authorship is 100% engineering. And a large share of the engineering backlog is work that was waiting on an engineer’s time, never on an engineer’s judgment.
If that’s what comes back, it points at the real constraint. Writing code was never the bottleneck for most of the roadmap. The bottleneck is that one team is the only path to production, and every change, however small, however clearly specified, has to pass through it. Make that team twice as fast and you’ve made the line move a little quicker. The line is still there.
The question for next quarter
Faster generation was the easy win, and most teams have taken it. The harder question is structural, and it’s about who is allowed to author a change that ships. As long as the answer is “engineers only,” the engineering backlog is bounded by engineering headcount, and roadmap velocity with it. No editor plugin changes that.
Run the audit. If the three numbers come back the way I’ve predicted, the next conversation to have is about the queue, and we’ve written that one up separately: Your team can build a feature in a day. Shipping it still takes a quarter.
Marketing Lead
Discover what the future of product teams looks like!
