AI Writes More Code. Why Doesn't Delivery Speed Up?
Episode hosts
What we discussed on the recording
The second 3 AImigo episode asks why AI speeds up an engineer but not end-to-end delivery. The Theory of Constraints shows that local optimization moves the queue into alignment, verification, or human attention. Flow starts from an accepted outcome and is measured by total cycle time.
A ten-person company exposed context through a tracker, memory, and incident tools, but human attention became the limit. Expert feedback moves into the short loop: involve a person by exception and turn each problem into a rule, prompt, or eval. The expert improves the mechanism instead of repairing outputs.
End-to-end work needs architecture, not an assistant in every product. Wardley Mapping informs what to buy; bounded contexts and Team Topologies define boundaries. Human UIs return huge JSON, so agents need compact contracts. A pilot still requires security, expertise, and an authorized sponsor.
Licenses are not process change, and resistance may be rational. Without an internal owner, old incentives return. Exploration begins with bounded hypotheses; operations are judged by outcomes. Tokens and AI-code share describe activity. Value appears in time to an accepted result, quality, rework, and business effect.