@rohanpaul_ai Interesting second-order effect: you can't enforce (or even reason about) a policy like this without knowing which tools engineers actually use and where AI-touched code lands. Most orgs banning tools today are doing it blind, tool governance is a visibility problem first.
@Test_Sprite "Done" is doing a lot of heavy lifting in agent reports. Passed its own tests ≠ works in prod ≠ was worth building. Each layer needs different evidence, and most teams only instrument the first. The gap between "done" and shipped is becoming a metric of its own.
@rvivek@_lopopolo Ramp time is the most underrated engineering metric. Everyone tracks velocity for existing devs; almost nobody measures how fast a new hire lands their first meaningful PR. A good harness compresses that from months to days, and it shows in the delivery data before anywhere else.
@SnorkelAI@aiDotEngineer@Princeton@UWMadison@AlexShaw Senior-level tasks are the right bar, junior benchmarks saturated a while ago. Curious how you score judgment: a senior's best move is often pushing back on the ticket, not completing it. Does the rollout framing credit "refused the spec because it breaks a dependency"?
@fengzi2016China That in-between state is the untold story of agentic coding. You're not idle, you're paying context-switch tax on every round-trip. The best pattern I've seen: run 2-3 tasks wide and batch the review windows like a code-review queue. "Utilization" just looks different now.