When part of the team is human and part is autonomous, velocity metrics lie, so the engineering manager's job becomes designing the operating model where people and agents ship together. An agent can produce a sprint's worth of code in an afternoon, which makes story points meaningless and moves the manager's leverage to review gates, evals, and cadence.
Engineering management ran on a familiar loop. Estimate the work, plan the sprint, track velocity, unblock the team, review the output, ship, repeat. The metrics were proxies for human effort, and they worked because humans wrote the code.
Then part of the team became autonomous. An agent can generate in an afternoon what used to take a sprint, which makes story points meaningless and lines of code actively misleading. A manager who keeps measuring velocity is measuring the wrong thing faster.
From managing effort to designing the operating model
In the AIDLC method, the manager's real job is the operating model that lets people and agents ship together. How does a spec flow from the product manager to the agent? Who reviews the generated diffs, and against what? What does the eval gate let through? Where do the humans spend their now-scarce attention?
The answer is consistent: humans move to judgment, agents handle volume. The senior engineer reviews instead of types. The QA engineer designs evals instead of writing assertions. The manager's job is to build the system where that division actually works, with clear review gates, a real eval suite, and a cadence tied to outcomes instead of activity.
The metrics that survive
Velocity proxies do not survive contact with agentic output. What survives is shipped behaviour that holds under load, regressions caught before release, and movement on the success metric the Frame phase defined. A manager who reorients the team around those numbers gets honest signal. One who clings to story points gets theater.
The planning horizon also compressed. When a slice ships in two weeks and the next is shaped by the traces of the last, long roadmaps become fiction. AIDLC runs in two-week vertical slices with a demo at the end of each, and the manager owns that rhythm.
If your team adopted agents but still runs 2022's sprint ritual and 2022's metrics, the ritual is now hiding your real progress instead of revealing it.
The engineering managers who win
They design the operating model, not just the sprint. They move their people to judgment and let agents take the volume. They measure outcomes, not activity. And they run a cadence short enough that the traces of the last slice shape the next.
Managing a team that adopted agents but kept the old rituals?
Most AI projects stall because nobody on the team knows how to design agents, manage token budgets, or wire production evals. I build that layer for B2B companies so the feature actually ships and keeps shipping.
Senior engineer turned AI specialist. React, Next.js, AWS, agent orchestration.
Direct collaboration across UAE, Europe, and US time zones.
Discovery, role design, MCP integration, evals, and production deployment.
If you want an operating model where humans and agents ship together cleanly, book a discovery call and we will design it for your team.