Use the rule you need. Keep the whole loop.

Name the affected user, observed friction, desired outcome, and unresolved question before selecting an AI capability.

Best for

Feature requests, roadmaps, workflow redesign

Point in flow

Before research or implementation

Try this

“Restate the user problem, separate evidence from assumptions, and identify the decision required before proposing a feature.”

Requirements, source authority, constraints, decisions, and evidence should live where future work can inspect them.

Best for

Long builds, multiple repositories, agent handoffs

Point in flow

Before and throughout delivery

Try this

“List the authoritative sources, conflicts, stale assumptions, and context that the next session must preserve.”

Explore competing explanations and tradeoffs before allowing the first plausible implementation to become the plan.

Best for

Ambiguous defects and high-cost commitments

Point in flow

Before choosing direction

Try this

“What would need to be true for each option to be correct, and what is the cheapest evidence that distinguishes them?”

Keep each session bounded around one decision, change, or verification loop that can be completed and handed off.

Best for

Agent sessions, complex backlogs, parallel work

Point in flow

At task definition

Try this

“Complete this bounded outcome, verify it, and stop before expanding into adjacent improvements.”

Define success, protected behavior, exclusions, failure boundaries, and how the result will be checked.

Best for

Implementation plans and acceptance criteria

Point in flow

Before mutation

Try this

“Turn this request into observable outcomes, explicit non-goals, protected behavior, and acceptance evidence.”

Name the sequence, interfaces, owners, risks, checks, and reversal path so implementation requires no hidden product decisions.

Best for

Multi-layer features and handoffs

Point in flow

After direction is approved

Try this

“Trace the approved change through its files, contracts, dependencies, tests, owners, and rollback sequence.”

The person or agent performing the change should not silently redefine intent, evidence, or approval.

Best for

AI coding agents and delegated teams

Point in flow

At assignment and review

Try this

“Implement the approved plan; surface any decision not covered by it and stop for direction.”

Complete one useful journey across every required layer before increasing feature breadth.

Best for

SaaS features, integrations, migrations

Point in flow

During implementation

Try this

“Identify the smallest end-to-end journey that provides value, preserves state, exposes failure, and can be reversed.”

Correctness, product intent, security, maintenance, accessibility, and risk require distinct review lenses.

Best for

Release candidates and consequential changes

Point in flow

Before readiness advances

Try this

“Review independently for product intent, correctness, security, user experience, regression risk, and unsupported claims.”

Walk the rendered user journey through ordinary, slow, denied, empty, invalid, interrupted, and recovery states.

Best for

Web, mobile, and AI interaction quality

Point in flow

After integration, before release

Try this

“Exercise the real task at supported sizes and input modes, including every material failure and recovery state.”

Convert root causes into a durable test, constraint, alert, operating rule, or product decision.

Best for

Production defects and repeated regressions

Point in flow

After surprises or failure

Try this

“Find the root cause and encode the narrowest durable check that would have prevented or detected it.”

Leave the next session an honest record of what changed, why, what was verified, what remains uncertain, and what comes next.

Best for

Long-lived work and agent coordination

Point in flow

At every handoff

Try this

“Separate completed work from assumptions and record changed files, decisions, evidence, unresolved risks, and the next task.”

Delegation does not transfer accountability for the problem, tradeoffs, evidence, risk, or decision to ship.

Best for

Prioritization, claims, ethics, release decisions

Point in flow

At every decision gate

Try this

“Present the case for and against release from observed evidence and unresolved risk. I will make the decision.”