Velocity Is a Weaker Story in the AI Era
Best for: VPs of Engineering, DevEx leaders, heads of delivery, product operations leaders, engineering managers, and executives who still rely heavily on activity metrics to explain software performance.
Use outside Forge: Very high. The article challenges output theater without attacking any specific framework.
Why this post matters now
For years, software organizations have struggled with a familiar measurement trap: the things that are easy to count are not always the things that matter most.
AI makes that trap worse.
When code, summaries, test cases, tickets, and documentation can all be produced faster, raw output becomes even less reliable as a leadership signal.
JetBrains' 2025 Developer Ecosystem research shows why. Developers now say non-technical factors such as collaboration, communication, and clarity are at least as important to performance as technical factors, and 66% do not believe current metrics reflect their true contributions. GitHub's enterprise survey found that developers often reinvest AI time savings into system design, collaboration, and meeting customer requirements more effectively. Sonar's 2026 survey adds a harder operational edge: AI output is rising quickly, but the bottleneck is shifting to verification. Stack Overflow's 2025 survey shows that more developers distrust AI accuracy than trust it, especially experienced developers in accountable roles.
This is the core measurement problem of the AI era.
Velocity can rise while confidence falls.
Why the old story breaks down
Activity metrics were always blunt. AI turns them into optical illusions.
A team can now show more visible motion with the same number of people:
- more code written or proposed
- more pull requests opened
- more test cases generated
- more documentation produced
- more items closed
But none of those necessarily tell you whether the team is making better decisions, reducing rework, improving release confidence, or delivering more value to the customer.
In fact, AI can increase some of those numbers precisely when the underlying system is becoming less healthy. A large body of generated output can create more review load, more subtle bugs, more ambiguity, and more hidden downstream cost.
That means velocity is not just incomplete. In some contexts, it becomes actively misleading.
What leaders should pay attention to instead
This does not mean measurement becomes impossible. It means measurement has to move closer to confidence, flow quality, and downstream value.
A better scorecard in the AI era usually asks questions like these:
Are we making the right work clearer earlier?
Upstream clarity matters more when output is cheap. If teams can generate more, then ambiguity gets more expensive, not less.
Are we reducing review friction or increasing it?
If AI speeds creation but slows review, the organization is not truly faster. It has just moved the queue.
Are engineers spending more time on higher-order work?
GitHub's research is useful here because it shows that time savings often move into system design and collaboration. That is a good sign - but only if the operating model actually protects that shift.
Are our metrics rewarding the right contribution?
JetBrains' data is a strong warning. If most developers do not believe the current metrics reflect their contribution, then leaders are likely over-reading surface activity and under-reading the work that holds delivery together.
Are we improving outcomes that survive downstream?
The best metrics are the ones that do not stop at the point of production. They follow the work into review, release, adoption, reliability, and customer impact.
What this means for engineering leadership
The leadership job is changing.
In an earlier phase of software delivery, managers often focused on how to increase output from scarce engineering time. In the AI phase, the harder question is how to increase quality-adjusted movement through the system.
That means leaders increasingly need to care about:
- time to confidence, not just time to draft
- decision quality, not just task completion
- review and validation capacity, not just generation capacity
- clarity of goals, not just backlog volume
- downstream rework, not just upstream motion
This is one reason Gartner is now talking about AI-native software engineering as a shift from implementation toward orchestration. As AI handles more of the task-level production work, human leadership moves toward problem framing, system design, risk management, and orchestration across the flow.
A more credible executive narrative
If you are reporting to a board, CEO, or CFO, “we are faster” is becoming a weaker claim unless you can also explain:
- faster at what
- with what level of confidence
- at what downstream cost
- with what effect on customer outcomes
- and with what risk posture
The stronger narrative is not raw velocity.
It is controlled acceleration.
It is the ability to move work with higher clarity, better review economics, stronger release confidence, and more time spent on genuinely high-value engineering decisions.
That is a story both technical and executive audiences can respect.
The strategic takeaway
AI does not make measurement less important.
It makes shallow measurement more dangerous.
Velocity still has a place. But as a standalone story, it is losing explanatory power.
The organizations that keep managing software through output optics will feel busy and still struggle to prove value.
The organizations that learn to measure clarity, confidence, flow quality, and downstream effect will understand what AI is really doing to their delivery system.
That is the better management advantage.
Selected references used in this draft
- JetBrains Research, The State of Developer Ecosystem 2025: Coding in the Age of AI, New Productivity Metrics, and Changing Realities (January 2026).
- GitHub Blog, Survey: The AI wave continues to grow on software development teams (August 2024; updated April 2025).
- Sonar, Sonar Data Reveals Critical “Verification Gap” in AI Coding: 96% Don't Fully Trust Output, Yet Only 48% Verify It (January 2026).
- Stack Overflow, 2025 Developer Survey - AI.
- Gartner, Top Strategic Trends in Software Engineering for 2025 and Beyond (July 2025).
Part of the AI-native delivery series on this blog.