👋 Hey {{first_name|there}},
Your dashboard says the team is shipping more since the AI rollout. Here's how to tell if that's real, and what it might be costing you when nobody's looking.
Why this matters
Picture the board deck. Commits are up. Story points up. There's a velocity line bending the right way ever since you switched on the AI tooling, and it photographs well. Then the director who used to run engineering somewhere leans in and asks the thing the slide can't answer. Are we getting more working software in front of customers, and is it holding up once it's there? Quiet. Because none of those numbers go anywhere near that question.
Most engineering dashboards have this problem. They count effort. Effort is easy to measure and easy to push up, and it got a lot easier the day a model could write a week of commits before lunch. The trouble is that effort and delivery aren't the same thing, and the metrics that look best on a slide are usually the ones furthest from whether anyone actually got value out of the work. You can be busier than you have ever been and shipping less than you were last year. I have watched a team do exactly that.
The last two issues built up to this. First, I argued that AI mostly shoves your bottleneck downstream, into review and integration, and the deploy step (Lesson #56). Then that pretending it didn't happen runs up what I called an instability tax, paid later in incidents and rework (Lesson #57). Enough diagnosis. This is the part where I hand you something you can actually put on a screen.
🧭 The shift
From: counting how much your engineers produce.
To: watching whether the work reaches production, quickly and in one piece.
None of this needs a new platform or a productivity suite. It mostly needs you to look at fewer things. DORA has spent more than a decade, and something north of thirty thousand survey responses, narrowing down to a handful of measures that actually track with how a company performs. The handy part is they fall into two camps, speed and stability, and you have to watch both. Most orgs report one and quietly fudge the other.
Three numbers get you most of the way there:
Lead time for changes. How long does a commit take to get to production? That's your flow, start to finish.
Review wait. This is the one the last two issues were really about. How long does a change sit there, opened and ignored, before anyone reviews it?
Change-failure rate. When you ship, how often does something break. This is what speed costs you when you aren't careful.