Status: Pilot approved · first teams up next
APEL is my Agentic Product Engineering Lifecycle. It takes its flow from Kanban: pull work when it’s ready and cap work in progress. For agent-assisted delivery, it adds a spec with test intent before any agent builds, a named person who approves each phase, and evidence before release.
Working rules
- Pull when ready. Ready + unblocked + capacity = eligible to pull. Product sets priority, and each engineer works one item at a time, two at most.
- Specify before building. Carry intended behavior and test intent into implementation.
- Keep decisions owned. Agents assist; people approve changes and accept outcomes.
- Verify the release. A successful build is only one checkpoint.
Track waiting, rework, and failures to improve the system. Staging runs on demand; production releases weekly. Teams can choose a different cadence without changing the model.
From Sprints to continuous flow
An engineer who finishes on Tuesday pulls the next eligible item that day. After validation it can reach staging on demand, then the weekly production window, without waiting for a Sprint to start. Resolve blocked work explicitly; starting more work does not make a dependency disappear.
Retiring Sprint events still leaves a need for shared goals, coordination, stakeholder feedback, and retrospectives. Make those responsibilities explicit, and pair effort measures with quality signals, as Goodhart’s Law reminds us.
The Scrum Guide requires neither velocity nor Sprint-linked releases. They’re common local practices, and they’re what APEL replaces.