Agent-planned sprints
Sprint planning starts from the specs, not a gut feel: PlanningAgent sizes the backlog, schedules it along the dependency critical path, and shows its reasoning, then a human commits.
How it works
Each item is sized (1/2/3/5/8) from its own requirements artifact, not a guess at a planning meeting.
A Critical Path Method pass computes earliest and latest start, slack, and the critical path across the dependency graph — sprints are packed in dependency order, and a dependency cycle surfaces as a named warning instead of a bogus schedule.
The agent orders items for delivery and explains why, in plain language a lead can push back on.
If the agent's output doesn't parse, planning falls back to a simple 1-point/backlog-order plan rather than blocking.
A plan only becomes real once a human commits it at the plan gate — and a committed plan can't silently change afterward.
The agents behind it
Human control
Plan gate — a human commits the sprint; freedom to reorder inside a sprint is real, but each item still passes its own implementation gates.