Your Team Got Faster. You Became the Queue.

AI made your team faster. The constraint moved to whoever approves the work, and that's you. Here's the data on the queue nobody in your company is measuring.

Quick question. Has your team's output gone up in the last eighteen months?

Almost certainly yes. More code, more drafts, more decks, more proposals. More of EVERYTHING! Everyone's got the tools now and everyone's moving.

Second question: is anything actually shipping faster?

Hmm. Not sure...

That gap between those two answers is what this article is about, and there's a very specific reason for it that almost nobody in your company is measuring.

The numbers are wild

I went looking for data on this because I kept hearing the same complaint from founders and I wanted to know if it was real. It's real, and it's worse than anyone's saying.

So, Faros AI published The AI Engineering Report 2026, built on two years of telemetry from 22,000 developers across more than 4,000 teams. Not a survey about how people feel. Actual measurement of what's happening across the software development lifecycle.

The top line looks brilliant. Task throughput up 34%. Epics completed per developer up 66%. PR merge rates climbing.

Now go one layer down.

Median time in code review: up 441.5%. Median time to first review up 156.6%. Average time spent in review up 199.6%.

They have a name for this and it's called The Acceleration Whiplash.

And it's not just slower, it's messier. The same dataset shows 861% more code churn and 242.7% more incidents per pull request, with no measurable improvement in organisation-level delivery metrics at all. Meanwhile 31% more pull requests are merging with zero review because reviewers can't keep up with the volume.

Sit with that one for a second. Your quality gate didn't get stricter. It stopped existing, and nobody sent a memo.

Why the review got so much harder

I'll include Anand and Wu's framework on how and when to use AI to explain this. So basically there are 4 approaches where some tasks are best handled by AI alone, others through human-AI collaboration, and some still require purely human judgment. One of those approaches is called Quality control, which has explicit data and high cost of errors. This is why when this approach is used, AI produces the work, and a human verifies it.

Now that his is clear, there's a detail in the Faros report that explains the whole thing, and it applies far beyond code.

AI-generated work is superficially convincing. It's well-structured, idiomatic, stylistically consistent, it looks like work produced by someone who knew exactly what they were doing. When the problems exist, they're underneath the surface. Catching them means reading properly, reasoning about intent, and reconstructing what the thing was supposed to do in the first place.

That is slow, expensive cognitive work. Considerably slower than scanning for obvious errors.

So the person doing the reviewing isn't just handling more volume. Remember, we are talking about high cost of error so each individual item now costs them more to check, more things to look at, and each one takes longer. You can see where this goes, right?

And it isn't only engineering

A 2026 survey of 2,078 workers found that 57% of managers and above have had to redo somebody else's AI-created work, compared to 38% of individual contributors. Above senior manager level, it goes past 60%. At companies that actually mandate AI use, 73% of workers have had to fix a colleague's AI output.

And 77% now say they review a coworker's work more carefully when they know AI was involved.

Research from BetterUp Labs and Stanford put a number on the cost of each incident: around two hours of rework, every time.

So it's not a code problem. Everything got faster to produce and slower to trust.

What makes it invisible then?

I read something in Forbes about this some time ago:

Fixing this kind of work produces no artifact.

The person who generated the flawed deck has a deck. It shows up in the record as output. The person who spent two hours fixing it has a document that now simply says what it should have said in the first place.

One of those people looks productive. The other one looks like they had a slow afternoon (unfortunately).

That's why nobody is measuring your queue. There's nothing to measure. The work you do doesn't leave a trace by design.

Your calendar shows meetings. Your Slack shows responses. Nothing anywhere in your company shows the six hours you spent last week being the last set of eyes on things that looked finished and weren't. And we both know that is the truth.

So how did you become the queue?

You never gave away the approval.

Think about what you actually handed over in the last two years. Tasks, mostly. The doing. The making. And your team took it, and then AI made them roughly a third faster at all of it.

But the approving and the final read? The "just send it to me before it goes out"? That stayed exactly where it always was - with you.

Their production rate went up 34%. Your review capacity went up by nothing, because there's still one of you and you still have the same number of hours, except each item now takes longer to check properly because it's convincing on the surface and unreliable underneath.

You are not slower than you were. The input rate changed and nobody redesigned the step you're standing in.

There's a study of over 100,000 GitHub developers that captures this perfectly. Coding activity from autonomous agents rose 180% at the commit level. By project level that had fallen to 50%. By actual releases, 30%. The researchers call it a weak-link production structure. Writing accelerates faster than the human stages needed to turn work into something shipped.

You are the weak link in position, not in ability.

What this is actually costing?

Your team learns to wait. When everything sits in your queue, people stop pushing and start parking. Work slows down at the edges long before it reaches you.

Your quality gate becomes theatre. Remember that 31% of pull requests merging with no review at all? That's what happens when a queue exceeds capacity. Things start going through unchecked, and the person most surprised will be you.

And the work only you can do stops happening. Every hour on the review pile is an hour not spent on the thing nobody else in the building can do. You've become the most expensive proofreader in your own company.

What actually fixes it

You can't work faster and you definetely cannot out-review this, the maths doesn't allow it.

Separate approval rights from task ownership. These are two different things and most founders have only ever delegated one of them. Make a list of every recurring approval that currently requires you. Then for each one, ask what would have to be true for it not to. Usually the answer is a documented standard, which brings us to the next one.

Get the standard out of your head. The reason everything comes back to you is that "good" only exists in one place, and it's inside your skull. If nobody else can tell whether something is right, they have no option but to ask you. That's not a discipline problem in your team. It's a documentation problem in your company.

Build review capacity that isn't you. Peer review between team members. A second pair of eyes that isn't the most senior pair. Categories of work where 80% is genuinely fine and everyone knows it.

Decide what genuinely needs a gate. Not everything does. The habit of checking everything was affordable when your team produced half as much. It isn't now, and pretending otherwise is how the gate collapses on its own.

The uncomfortable version

Your team got faster and you didn't restructure around it. That's it. That's the whole diagnosis.

And the reason it hasn't been fixed isn't that you don't know it's happening. You know. You just haven't reflected on it because you've been in the queue.

I work one-to-one with founders on exactly this: getting the decisions and the approvals off your desk permanently, so the business stops running at the speed of your availability. Not more delegation of tasks. The authority underneath them.

I'm taking three founding clients at the moment, and I'd rather have a conversation and tell you it's not the right fit than sell you something that isn't.

if you want to talk or work with me, go to my contact page and book your call.

Anyway, three things to sit with this week:

How many separate things are sitting somewhere right now, finished, waiting only for you to look at them?

Which of those genuinely required your judgement, and which required your permission?

If your team's output went up another 30% next quarter, what exactly in your company would absorb it?