Your teams are writing far more code than they did two years ago but you're probably not seeing a lot more valuable outcomes.
McKinsey put numbers on it in its Technology Trends Outlook 2026: "AI tools increased coding activity by 180 percent, but shipped releases rose by only 30 percent."
Read that again as a delivery leader. Coding activity nearly tripled. Releases went up by less than a third. Roughly 150 points of new output went somewhere other than production.
If you bought AI coding tools and can't show that delivery improved, this is your situation, and you are not alone in it. The same report says that in 30 percent of companies, productivity fell after teams started using agentic AI tools.
So where did the other 150% go?
It went into the queue
Code is not software until someone has reviewed it, tested it, integrated it, and decided to ship it. AI made the first step fast. It did very little for the rest.
In my work with teams using AI to build software, the extra output lands in four places.
Review. A developer with an AI assistant can open several pull requests in the time it used to take to open one. The people reviewing them did not get faster. Either the queue grows, or reviews get thinner. Thinner reviews cost you later.
Test. More code needs more testing. If the test suite, the test environments and the test data were sized for the old pace, they are now the constraint. AI-written tests help, but someone still has to trust them.
Integration and deploy. Many organizations still release on a calendar: a change board, a release train, a monthly window. Triple the code going in and the window stays the same size. The work waits.
Decisions. This is the one leaders miss. Every story needs someone to say what should (and should NOT) be built, whether it is right, and whether it is done. Those decisions take the same time they always did. When code arrives faster than decisions, teams either wait or guess. Guessing often simply builds the wrong thing quickly.
McKinsey's own diagnosis is close to this: "AI is producing code faster than human systems can review, test, and deploy it securely."
I watched this play out with one team. They were already having trouble getting pull requests reviewed before AI arrived. Once AI was writing the code, the queue grew, and we accepted more risk by reviewing less thoroughly. The result was no surprise: as soon as the new code reached the QA environment for user acceptance testing, there was an immediate influx of new defects.
Writing code was never the constraint
None of this is new. For most teams, typing the code was a small share of the time between an idea and a release. The rest was waiting: for clarity, for a review, for an environment, for a decision, for the release window.
Speed up a step that is not the constraint and you don't get more out the other end. You get a bigger pile in front of the step that is. AI coding tools did exactly that, at a scale nobody had seen before.
That is why the gains are so uneven. McKinsey reports that "80 percent of software engineers using AI tools see a productivity acceleration of around 3 percent, the top 20 percent see a 55 percent boost." Same tools, very different results. The difference is the system around the developer.
It is also why I don't accept the claim that AI made agile obsolete. Small batches, a visible queue, a team that owns its work through to release, and a regular look at how the work is flowing: those are the practices that find and remove constraints. Teams that have them can absorb faster code. Teams that don't are the ones getting slower.
Four questions to ask this week
You can find your own 150% without a consultant. Ask your teams:
- How long does a finished change wait for review? Compare it with a year ago.
- How long from "code complete" to production? If that number didn't move, the tools didn't change delivery.
- Who decides what a story means, and how long does that answer take? Look for work that started before anyone was sure.
- What do we measure? If the dashboard shows lines of code, pull requests or AI usage, it is measuring the 180. Measure the 30: releases, cycle time for a story, and defects that reach customers.
The answers will point at one or two places where work piles up. That is where to spend the next quarter, not on another tool license.
The short version
AI gave your teams a faster way to write code. It did not give them a faster way to deliver. The gap between 180 and 30 is your delivery system, and it can be fixed.
On Wednesday, October 14, I'm taking this further in a free webinar: "You Bought the AI Tools. Why Isn't Anything Getting Better?" Register here. If you lead Scrum Masters or coaches, bring them.
Source: McKinsey & Company, Technology Trends Outlook 2026 (September 2026), "Agentic software development," p. 10. Report summary