An engineer shipping 30% more PRs doesn't mean your organization is delivering 30% more value. We just wrapped our State of AI SDLC digital summit, where we went deep on how technical leaders are adapting for the AI era. 3 big takeaways we heard: 📌 Context is the leverage. AI agents need your specific services, dependencies, and past decisions to move beyond generic output. 📌 Activity ≠ impact. Don’t just track token spend or PR volume. Measure whether AI is unlocking capacity for harder problems. 📌 Systems over tools. You can’t bolt agents onto legacy processes. The orgs that win will redesign how work flows to keep humans in the loop. Watch the full summit on demand to see how our CEO & Co-Founder Mike Cannon-Brookes and leaders from Vercel, Lovable, Dropbox, 1Password, and DX are building the new AI playbook. Link in comments. ⬇️
AI productivity isn’t about doing more work. It is about doing harder work optimized way.
"Activity ≠ impact" is the one I'd put on every leadership dashboard. From what I see with clients, the teams getting real value from AI agents aren't the ones with the most tools. They're the ones whose Jira and Confluence were already well structured: clear issues, documented decisions, connected work. That's the context agents actually feed on. AI doesn't fix messy processes. It scales them.
PR volume is also easy to split before you trust it. Tag AI-assisted PRs, then compare change failure rate (one of the DORA delivery metrics) and how many get reverted or hotfixed within a week against the rest. If the extra PRs come back as rollbacks, the capacity was borrowed from next sprint.
Measuring AI adoption through activity alone can miss the bigger picture. Focusing on context, meaningful outcomes and redesigned workflows gives teams a clearer view of whether AI is actually improving how work gets done.
Measuring AI adoption through activity alone completely misses the enterprise reality. The shift from tracking token spend or PR volume to measuring actual capacity unlocked is the defining difference between an AI experiment and a strategic deployment. As organizations transition to multi-agent workflows, the "activity ≠ impact" principle is critical. An engineer shipping 30% more PRs only creates value if the business operations can actually absorb and commercialize that output. Otherwise, we are just generating code inventory that sits in a bottleneck. For the engineering leaders driving this shift: If PR volume is a vanity metric, what telemetry are you using to prove these agents are actually driving sustainable business velocity?
There's a lot of context behind those PRs that truly show the health of the team. When leveraging AI to deliver PRs, I've seen teams encounter massive growth in lines of code (stay with me, I know that one makes most scoff). The AI doesn't always understand their source code and produces a more complex solution than someone who's familiar could have. In turn, your Reviewers and Approvers are now stuck trying to decipher a different language in much longer (and much more cases as you stated) samples. This led to longer Review/Approval times, and also elevated Revert Rates. It's important to identify true signals of success within the metrics if your tooling allows you to. Those signals tell the real story!
Activity =/= impact. Please send this formula to all organizations still counting employees hours instead of value. If it's right for agent, it's right for human. For years that organizations can't measure impact of their employees, now they need to measure impact of AI. Against what? It has to come against impact of humans, otherwise you can't tell whether a human should do that or AI. This is going to be a long journey of educating managements.
The “activity ≠ impact” point resonates. We’ve seen AI speed up code creation while the bottleneck moves to review, coordination, or simply deciding what should be built. Counting more PRs won’t tell us whether those problems are getting better. What makes the Atlassian ecosystem interesting here is that Jira already holds the context for the work. At Tempo, we’ve spent years helping customers connect people’s effort to that work; we’re now applying the same thinking to agents and AI spend. Together, those signals give teams a better starting point for understanding what AI is actually changing in how they deliver. Atlassian and its ecosystem are well suited for enterprises to usher in the new way work is planned and delivered.
Catch the full State of AI SDLC summit on demand 👉 go.atlss.in/132c3r