Skip to content
AI-Native PM

Frameworks & execution models

A model never gives you the same answer twice, and a framework is what makes the decision come out right anyway.

A framework does for product judgment what a good recipe does for cooking: a result you can count on even as the inputs keep changing. So on every AI decision we run one cycle, Shape · Ship · Track, then run it again until it holds.

The Shape, Ship, Track loop around your AI productA loop of three nodes drawn as a line diagram. Shape decides its behavior, Ship checks before it speaks, Track measures real use, and the loop returns to Shape. Your AI product sits in the center.ShapeShipTrackdecide its behaviorcheck before it speaksmeasure real useyour AI product

Framework Summary

Shape · Ship · Track

Every activity and artifact.

The Playbook
Shape

What should your AI product do?

Frame the problem

Decide what the model should do, and what good looks like, before you build.

Discovery guideProblem frameDefinition of good
Write the behavior, then the prompt

Put the behavior in a contract, then build the system prompt that enforces it.

Behavior contractStarter system promptEscalation rules
Choose the model and grounding

Pick the model as a product decision, then decide how it gets facts it was not trained on.

Model rationaleContext architectureGrounding decision
Prototype it

Build a real version yourself in a day, and let it find the cases your spec missed.

Working prototypePrototyping checklistEdge-case list
Ship

Can you prove your AI product does it well?

Build the guardrails

Decide what the model may never do on its own, and enforce it.

Input and output guardrailsEscalation pathRollout checklist
Build the eval suite

Measure that it works, because you cannot catch quiet failures by eye.

Regression evalsPass and fail thresholdsRelease gate
Earn trust when unsure

Design the uncertain moments instead of hiding them.

Preview and undo designSource and confidence display
Set the cost and speed budget

A good response is relevant, consistent, appropriate, affordable, and fast. You cannot maximize all five.

Latency and cost budgetRouting and caching plan
Track

Does your AI product still do it well?

Watch behavior in production

Measure what matters once real people are using it, and read real sessions.

Quality dashboardMetrics taxonomySession-review habit
Catch the drift

The model moves under you, so watch for slow decay.

Drift alertsRegression gate on model changes
Feed it back into Shape

Turn every failure you catch into the next turn's work.

New eval casesRanked list of contract fixes

Then you run the cycle again.

Continuous Operations

The umbrella across every turn of the cycle.

Govern the knowledge

A retrieval system is only as good as what it is allowed to read.

Curation policyRefresh cadenceConflict rules
Govern access and safety

Tell the system what each person is allowed to see.

Access-as-behavior rulesSafe-refusal patterns
Supervise the agents

Once a product acts on its own, every weakness compounds.

Supervision designIteration capsReliability budget
Build the team

A cycle only you can run is not yet a practice.

Hiring rubricOrg change plan

Behind the Frameworks

Product management tools created for the full AI product lifecycle.

The Discipline: what changed about the workA before-and-after wipe. A bright divider sweeps left to right. As it passes each row, the muted old way crossfades into the vivid new way and an accent underline draws in. A spec of features becomes a spec of behavior. Owned across functions becomes PM accountable for behavior. A known menu becomes new problems and new solutions. Acceleration is tablestakes. Owning this work is the job.WHAT CHANGEDBEFOREAFTERA spec of featuresA spec of behaviorOwned across functionsPM accountable for behaviorA known menuNew problems and new solutionsAcceleration is tablestakes. Owning this work is the job.

Using AI to move faster is becoming table stakes. Building with AI is a separate craft, and the work itself changes. What you ship becomes a spec of behavior rather than a set of features. You own how the model behaves when no other function does. And the menu of what is worth building is new, so the only way to know is to build it.

  • A spec of behavior
  • Owning the behavior
  • New problems to solve
Learn and practice the new discipline12 min read

The Human FactorsConcentric rings like a radar of the mind, from the outer ring Perception inward through Working Memory and Mental Models to the innermost Metacognition, with a human mind mark at the center. A pulse ripples inward through the layers; each faculty lights up in turn with its design recommendation.DESIGNING FOR THE HUMANPERCEPTIONWORKING MEMORYMENTAL MODELSMETACOGNITIONTHE HUMAN MINDPERCEPTION"Make the warning impossible to miss."

The part of the system the model has no access to, and the part the PM is there to protect.

The products that win will be the ones designed for the mind that has to use them, with its real and well-documented limits, not for an idealized user who has none. That mind is on the other side of every model. It is the part of the system the model cannot see, and the part the PM is there to protect.

  • Perception
  • Working memory
  • Mental models
  • Metacognition
Learn practical design principles16 min read

Shape · Ship · TrackA continuous cycle of Shape, Ship, and Track, then Shape again, because a probabilistic system is never finished.A probabilistic modelmade reliable1ShapeDecide how it behaves2ShipPut it behind guardrails3TrackCatch what users won't report

The whole practice. You shape how a model behaves, you ship it to a human behind guardrails, and you track whether it holds. Then you do it again, because a probabilistic system is never finished.

  • Shape
  • Ship
  • Track
Learn a powerful AI development cycle7 min read

The Playbook: Shape · Ship · Track, operationalizedA blueprint of columns headed Shape, Ship and Track, under one umbrella band labelled Continuous Operations, the umbrella across every turn of the cycle. Shape opens into Frame the problem, Write the behavior, Choose the model, and Prototype it. Ship opens into Build the guardrails, Build the eval suite, Earn trust when unsure, and Set the cost and speed budget. Track opens into Watch in production, Catch the drift, and Feed back into Shape. The rows draw in column by column, then a dotted feedback path runs from Feed back into Shape back up to the Shape column. The cycle opens into its activities. Continuous Operations runs across every turn.THE CYCLE, OPERATIONALIZEDCONTINUOUS OPERATIONSthe umbrella across every turn of the cycleShapeFrame the problemWrite the behaviorChoose the modelPrototype itShipBuild the guardrailsBuild the eval suiteEarn trust when unsureSet the cost & speed budgetTrackWatch in productionCatch the driftFeed back into ShapeThe cycle opens into its activities. ContinuousOperations runs across every turn.

The cycle, made operational. Each move opens into the activities you actually do, with Continuous Operations running across all of them, and every activity produces something real you can hand off.

  • Shape
  • Ship
  • Track
  • Continuous Operations
Get detailed steps to run the cycle14 min read

Go hands-on

Three levels of learnings beyond the frameworks

The Builder's Stack is the hands-on course where you take everything on this page and turn it into a product that behaves. It carries you from the fundamentals to the frontier.

Open the Builder's Stack
Level 1The FundamentalsLearn how software works, then direct an AI build of your own from idea to first real users.Level 2The PracticeMake a working build safe, tested, and ready for real users.Level 3The FrontierTake on the newest AI problems, from agent fleets to agent users.