AI adoption in engineering is settled. In GitKraken’s 2026 State of AI in Engineering survey of 554 developers and engineering leaders, 96% of teams had adopted AI coding tools and 84% of developers said they feel more productive. Only about one in five can actually measure the impact. 39% of organizations have no way to measure it at all, and another third rely on developers’ self-reports.

That gap is getting harder to live with. In nine months, the share of developers who work primarily through autonomous agents roughly quadrupled, from 7.6% to 28%. A third now keep agents running the entire workday, and nearly half do in the enterprise. Agents are turning into production infrastructure that writes, reviews, and ships code all day, and AI spend has become a real line item. Boards and CFOs want to know what it returns. “It feels faster” doesn’t answer that, and neither do lines of code.

This session is for engineering leaders who are being asked to justify AI spend, and the senior engineers who end up building the dashboards to do it. It has two parts.

First, the research: how agent maturity tracks with productivity, why the teams furthest along with agents are over three times as likely to track productivity metrics, what tool choice reveals about a team’s maturity, and the four practices that separate teams who can show results from teams who can’t.

Second, a live demo of GitKraken Insights, working through the questions leaders get asked most:

Where is AI working, by team, tool, and model, and where isn’t it?
What does each tool and model cost per unit of delivered output, and how much premium-model spend is going to routine work?
Are quality and delivery (DORA, cycle time, review load) holding up as AI-assisted changes grow?
Which repositories are ready for agents, and which aren’t?
How do you give every developer their own view of the data, so measurement helps them improve instead of grading them?

Along the way, we’ll share what we learned measuring GitKraken’s own engineering org first (2.53x output in six months), and how customers apply the same approach. That includes a 600-developer health plan that rolled out in under a day, and Case IQ, whose CEO put the problem plainly: “When I sit down with our board, they don’t want to hear that developers ‘feel’ more productive. They want to see the data.”