AI & Tech

AI Writes More Code. Why Aren't You Shipping Faster?

Your team writes code faster with AI, but customers aren't seeing more. The constraint moved to deciding what to build and getting it safely to users.

Eduardo Diaz · 2026-09-25 · 6 min read

AI coding product discovery software delivery code review

A question I hear from founders a lot right now goes something like this: "My engineers all use AI. They say they're writing code faster than ever. So why does it still take us a month to get anything in front of customers?"

It's a fair question, and big companies are asking it too. Last week, TNW reported that Oracle rolled out ChatGPT Enterprise and OpenAI's Codex in April and May, and that within three months 80% of staff were using the tools. The same report says this hasn't sped up delivery to customers. Co-CEO Clay Magouyrk put it plainly: writing code faster "doesn't mean that suddenly everything is 1000 times faster." Oracle is still reworking its testing, validation, deployment and release processes to catch up.

If a company with Oracle's resources hits that wall, your ten-person team will too. The good news is that the wall is predictable, and you can plan for it.

Typing was never the slow part

When I think back over two decades of building products in analytics and healthcare, I can't name a project that was late because people couldn't type code fast enough. Projects were late because we built the wrong thing first, because requirements changed halfway through, because testing found problems at the end, or because the release needed sign-off from three people who were busy.

AI made the typing part cheap. It didn't touch the rest.

So the constraint moved. It moved upstream, to deciding what to build. And it moved downstream, to everything that has to happen between "the code exists" and "a customer is getting value from it."

What the research says

The data on this is more consistent than the hype suggests.

Google's 2025 DORA report, based on survey responses from nearly 5,000 technology professionals, found that 90% of respondents use AI at work and more than 80% believe it has increased their productivity. It also found a positive relationship between AI adoption and delivery throughput. But AI adoption still had a negative relationship with delivery stability. Their summary is the line I keep quoting to clients: "AI doesn't fix a team; it amplifies what's already there."

The review problem shows up in this year's surveys too. GitLab's AI Accountability Report, run by The Harris Poll with 1,528 developers and technology buyers, found that 85% agree AI has shifted the bottleneck from writing code to reviewing and validating it. And 80% said their organization adopted AI tools faster than it developed policies to govern them. Just this week, Qodo's 2026 State of AI Code Quality report (a Censuswide survey of 500 US developers and 300 engineering leaders) found both groups named reviewing and validating AI-generated code as their top delivery constraint, and 89% of organizations reported an AI-related production incident.

Then there's the gap between how fast AI feels and how fast it is. In METR's 2025 randomized trial, 16 experienced open-source developers took 19% longer on tasks when AI was allowed, while believing afterward that it had sped them up by 20%. To be fair, METR now calls those results out of date. Its February 2026 update on late-2025 tools points toward a speedup, though the authors call their own data "very weak evidence" for its size. I don't read either study as "AI doesn't help." I read them as a warning that your team's sense of speed isn't a measurement.

A few of these surveys come from vendors who sell review tools, so I hold the exact percentages loosely. The direction is what matters, and every source points the same way.

Upstream: deciding what to build

Here's the part that gets less attention. When building gets cheaper, the cost of building the wrong thing doesn't go down. It goes up, because you can now build much more of the wrong thing before anyone notices.

DORA makes a related point: AI is most useful when it's pointed at a clear problem, and a user-centric focus amplifies its positive effect. That's what product discovery is for. Talking to users, testing assumptions, and cutting scope down to the one workflow that proves the idea. None of that got faster with AI. If anything, it matters more, because the queue of things you could build is now effectively infinite.

Downstream: review, testing, release, adoption

On the other side, every line AI writes still has to be read by someone who understands the system. It has to be tested, deployed and monitored. Then a real person has to find the feature, understand it and change their behavior. That last step is where a lot of "shipped" work quietly dies.

If your team produces twice as many pull requests and your review capacity stays flat, you haven't doubled output. You've built a queue.

Where to look in your own process

If you want to find your real bottleneck, skip the opinions and look at where work waits. A few places I'd check:

Pick the stage where work sits longest and fix that one first. Speeding up anything else just moves the pile.

Why this is a team design problem

The teams I see getting real value from AI aren't the ones with the most tools. They're the ones where the people deciding what to build, the people designing it and the people engineering it work as one unit, with review and testing treated as part of the work rather than a gate at the end.

That's a big part of why we built AetherCraft as a product studio that owns strategy, design and engineering together. When discovery and delivery live in different places, every handoff is another queue. When they live together, a fast-moving AI-assisted team can actually turn that speed into a product in customers' hands. It's also how a focused MVP can realistically land in 6 to 10 weeks: the time savings come from deciding well and reviewing carefully, with AI handling more of the typing in between.

If your team is writing more code than ever and customers still aren't seeing the difference, I'd be glad to talk through where your work is waiting. Often the fix is smaller than it looks once you can see it.

Sources

Keep reading

← Back to Insights