Responsibilities, agents, and tools support the same connected system: build reliably, evaluate impact, and improve how you work. Choose the question you want to clarify.
Five human responsibilities, six agent profiles, and their boundaries.
BUILDImplement small, agreed changes with clear criteria and human ownership.
BUILD · VALUE · LEARNINGFlow Check, Refinement, Spec Clinic, Acceptance Slot, Value Review, and learning rituals.
BUILD · VALUE · LEARNINGPitches, specs, small tickets, and agent tasks within the complete loop.
BUILD · VALUE · LEARNINGConnect shared direction with autonomous teams across streams.
BUILD → VALUE → LEARNINGEmbed monitoring, alerting, and resilience in specs, acceptance, and learning.
BUILD · LEARNINGDistinguish reasoning effort, capability, and autonomy.
BUILD · VALUE · LEARNINGAll steps, handoffs, evidence, and feedback connections in the explorer.
BUILD · VALUE · LEARNINGUse templates, specs, prompts, and the scorecard.
Start with an agreed, versioned spec. Give agents bounded tasks with clear evidence.
Starting points for a pilot. Decisions can also follow new findings between rituals.
PO and team clarify investment limits with the responsible people. These may be set by company strategy, the board or a higher-level initiative. The limits bound the investment; they do not promise every desired feature within that investment.
Agree on human effort, AI costs, expected benefit and checkpoints. Consider capacity and calendar duration separately.
Example assumption: at most 5 team working days and a €50 AI budget for the first useful cart. Review at halfway consumption.
Postpone convenience, variants or additional features; simplify workflows and shape smaller deliveries. Feature Partner and PO confirm the changed spec.
Price correctness, stock limits, security and necessary tests remain binding. The budget is not silently increased.
Continue with smaller scope or stop within the mandate; have the responsible authority explicitly confirm new limits. The decision, reason and owner remain visible.
The Value Loop then checks the benefit of the released change. Its measurement period is agreed separately.
The depth of clarification depends on risk and uncertainty. A small task needs a clear brief and evidence – often on a single work card.
Reproduce the failure, reference the agreed expected behaviour, add a regression test, fix and review. If the expected behaviour is unclear, briefly clarify it with the Feature Partner.
Value: Check recovery using failure or operational data. This usually does not require a new Feature Scorecard.
Record the desired outcome, owner, budget and acceptance criterion on the work card. Pull work within the available work limit; review according to risk and complete it.
Example: A help text correction needs a visual check; a small permission change also needs targeted access tests.
The incident owner coordinates diagnosis, containment and recovery. Urgent work visibly takes priority; identify the work being paused.
Keep required approvals and rollback arrangements clear. Complete evidence, the spec and regression tests after stabilisation.
The team uses experience to allow for support and operational work in its planning, limits parallel work and reviews the arrangement regularly. A new urgent task visibly displaces another; it is not silently promised on top. If scope grows, reprioritise or split it into a separate change.
Flow Check: ongoing work and blockers. Spec Clinic: a specific uncertainty or budget pressure. Value Review every 4 weeks: impact and trends. Recurring individual failures feed into Learning.
Template for an operational work card · View the example timeline