Showing all 6 practices. Answer the Diagnose questions and the most relevant come first.
Run small experiments on recurring problems
Treat recurring problems and performance gaps as chances to learn about the system. Frame each improvement as a small experiment you can test.
Start by asking your team: How quickly can we test and learn from an idea meant to fix a recurring problem?
Avoid · You’ll know it’s working when
Avoid
Treating retrospectives or lessons-learned sessions as improvement by themselves
Launching large transformation programs without testable hypotheses
You’ll know it’s working when
Problems raised in retros get fixed and stay fixed, and improvement actions get finished and checked.
Prove a change works before standardizing
Measure whether a change improves the outcome it was meant to before you standardize it. Close the loop by tracking outcomes instead of counting actions.
Start by asking your team: What have we learned about how the system is actually behaving, and what did we change based on that evidence?
Avoid · You’ll know it’s working when
Avoid
Introducing frequent change without measuring whether outcomes actually improve
You’ll know it’s working when
Every process change comes with evidence that it moved the outcome it targeted.
Retire what stopped working, keep what works
Stop or revise practices that no longer produce useful results. Build what does work into your standard ways of working.
Start by asking your team: Which practices should we stop, revise, or standardize based on what we've learned?
Avoid · You’ll know it’s working when
Avoid
Preserving outdated practices because they worked in the past
You’ll know it’s working when
Practices get dropped once the conditions behind them change, and lessons from one team show up in shared standards.
Bring AI decisions into retros and reviews
Put AI-influenced decisions on the table in retrospectives and reviews. Link outcome metrics back to the AI behavior or recommendations that shaped them.
Start by asking your team: How did AI influence this decision or outcome, and what patterns show up across AI-supported decisions?
Avoid · You’ll know it’s working when
Avoid
Treating AI as a passive tool rather than an active contributor to outcomes
Assuming AI self-optimization means the delivery system is learning
You’ll know it’s working when
Retros examine agent influence, and outcomes can be traced to the agent behavior behind them.
Monitor agent behavior and name an owner
Track agent behavior patterns over time with dedicated evaluation and monitoring loops. Give someone clear ownership for acting on what they surface.
Start by asking your team: What feedback actually reaches the model or agent configuration, not just the team?
Avoid · You’ll know it’s working when
Avoid
Reviewing individual AI outputs without examining behavioral patterns over time
You’ll know it’s working when
Drift and bias show up in monitoring before customers feel them, and someone acts on it.
Revisit the system when conditions change
Check whether changes in technology, teams, customers or dependencies mean the system has to evolve, including agent configurations.
Start by asking your team: As AI becomes part of delivery, how are lessons changing agent behavior, configuration, and governance?
Avoid · You’ll know it’s working when
Avoid
Improving human workflows while leaving automated failure patterns untouched
You’ll know it’s working when
Old practices and agent setups get revisited when the conditions behind them change, and faster delivery shows up in outcomes.
Overview
Principle intent
Continuously learn from system behavior, human and AI alike, and use that learning to improve how work is designed, delivered, governed, and supported.
Read the full intent
Delivery systems accumulate friction, outdated practices, and new failure modes as conditions change. Sustained performance depends on the system's ability to observe, learn, adapt, and retain what works. AI-influenced decisions have to be part of that learning loop just as much as human ones. Nothing exempts them from it by default.
When it holds: what it does for you
Problems stay fixed
Recurring issues get tested with small experiments and removed, so teams stop working around them.
Changes prove themselves
Only changes that move the intended outcome make it into standard ways of working.
AI inside the loop
Agent behavior gets reviewed and corrected alongside human decisions, so drift shows up before customers feel it.
When it is ignored: from early signs to a lasting condition
Early warning signs · Visible now, cheap to fix
Improvement actions get identified but are rarely finished or checked.
Process changes go in without evidence that they helped.
Old practices continue out of habit after the conditions behind them have changed.
What happens over time · Accumulates, costly to reverse
Known problems become fixtures that teams work around instead of removing.
Outdated practices stay embedded, and AI now repeats them faster.
Lessons stay local and leave when the people who learned them leave.
Intent Drift: The team built exactly what was asked. It is no longer what is needed.
Implementation Drift: The agents are building. Nobody is sure it still matches the plan.
Oversight Erosion: The review process still runs. It stopped meaning anything.
Take your Diagnosis to the next level by invoking AI Diagnostic Coaching tool.
What else have you observed? Optional. Shared only if you choose to start coaching.
Start a Diagnosis session: review and edit a draft containing your note, conditions you recognized in Step 3 (including whether they match partly), warning signs and consequence ratings before sending it. No scores or principle files are sent.