Competencies and Career Strategy
A practical 10-competency model with observable behaviours, an interactive profile tool and concrete career strategy as AI reshapes product management.

On this page6 sections
Two product managers apply for the same role. One runs sharp customer discovery and aligns sceptical executives cleanly, yet has never inspected a model trace or written an evaluation rubric. The other builds slick working prototypes in an afternoon, yet cannot explain which workflow constraint would make a customer pay for one.
Neither profile is universally "better." Different roles demand very different shapes, and strong product teams assemble complementary coverage rather than hunting for mythical unicorns who score maximum marks across every axis.
Use the ten-competency framework below to diagnose real strengths and gaps, map team coverage and plan deliberate career moves as AI compresses parts of the traditional PM job while raising the ceiling on others.
The ten competencies
| Competency | Core question it answers | Observable evidence |
|---|---|---|
| Customer and market understanding | Whose problem is this, and why does it matter now? | Observed workflows, workarounds, urgency signals and demand evidence |
| Strategy and commercial judgement | Which bet deserves capacity, and do the economics hold? | Explicit capacity trade-offs, full workflow unit-cost model and clear non-goals |
| Discovery and experimentation | What must we prove before committing engineering? | Falsifiable bet card, right-sized test, contradictory evidence and a changed decision |
| Product definition and experience | Can users understand, verify and trust the workflow? | Clear framing, interaction test on real failures, visible state and recovery design |
| Model and system fluency | How does the harness produce outputs or execute actions? | Trace walkthrough isolating source, retrieval, prompt, tool and model layers |
| Evaluation and quality | How do we prove the system clears the bar on edge cases? | Representative dataset, binary rubric, slice analysis and trace-to-eval loop |
| Building and technical collaboration | What can we make to learn, and where does specialist depth take over? | Bounded prototype with stated limits, sharp technical critique or fixed failure |
| Operating judgement and risk | What may the system do, and how do we recover when it breaks? | Authority boundaries, review staffing maths, kill switch and operating agreement |
| Delivery and influence | Can the organisation decide cleanly and ship through friction? | Decision brief, resolved disagreement, staged rollout and operational follow-through |
| Team leadership and learning | Are the right capabilities, context systems and habits in place? | Shared context/skills infrastructure, coaching, smart hiring and reduced review drag |
These axes overlap in real work. One customer observation often reshapes the problem framing, adds three exception cases to the eval suite and alters the pricing model. Treat the list as a diagnostic lens, not ten bureaucratic silos.
Know the working technical baseline
Every product manager building with AI needs a practical technical floor. Training models or owning production infrastructure remains engineering's job; understanding how the system actually works does not.
For the workflows you own, aim to do five things reliably:
- Explain the execution path: Describe what enters the model, how retrieval works, which tools exist, where state lives and where permissions are enforced.
- Read a failure trace: Run representative inputs, inspect the trace alongside an engineer and distinguish stale context, broken search, bad instructions, tool errors and model limits.
- Co-author the quality bar: Curate real cases and ugly edge cases with domain experts, resolve rubric disagreements and read sliced eval results critically.
- Model operational trade-offs: Quantify latency, review load, error blast radius, unit cost and recovery paths before expanding autonomy.
- Build small artefacts to learn: Prototype an interaction, test a prompt variation or run an evidence analysis yourself while stating plainly what is simulated and what requires production engineering.
Depth scales with consequence. A PM adding internal research clustering needs far less depth in permissions and recovery than one owning an autonomous billing or booking agent.
Rate observed behaviour, never titles or confidence
Assess each competency across four observable levels, plus not yet observed when you lack real evidence (lack of opportunity is not lack of ability):
| Level | Demonstrated behaviour |
|---|---|
| 1. Developing | Understands core concepts and contributes with guidance; needs help spotting hidden gaps or picking the right method. |
| 2. Working | Handles representative tasks independently, explains trade-offs clearly, checks results and pulls in specialists at the right boundary. |
| 3. Leading | Navigates high ambiguity, resolves cross-functional trade-offs, raises the team's quality bar and coaches peers. |
| 4. Shaping | Establishes durable practices and shared systems across multiple teams, sets hard boundaries and develops other leaders. |
Senior titles do not automatically mean level 4 across the board. A seasoned VP can sit at level 4 in commercial strategy and level 1 in trace debugging; an IC staff builder can shape evaluation practice across an entire org without managing headcount.
What progression looks like in practice
| Level | Model & system fluency | Evaluation & quality | Operating judgement |
|---|---|---|---|
| Developing | Traces one request through context, model and tool with an engineer; spots a point of uncertainty. | Reviews sample outputs with a peer and articulates why an answer fails. | Identifies a consequential action in a workflow and asks who can approve or stop it. |
| Working | Reads a live trace, isolates plausible root causes and tests a fix cleanly with engineering. | Curates representative cases, defines clear rubric checks and uses sliced results for a pilot call. | Defines permitted actions, review load, handoff and recovery with system owners. |
| Leading | Compares harness, routing and model trade-offs on quality, latency and cost to steer architecture. | Calibrates rubrics with domain experts, closes coverage gaps and wires trace-to-eval loops. | Sets proportionate rollout gates for ambiguous workflows with owned monitoring and kill switches. |
| Shaping | Establishes org-wide observability, context standards and technical decision practices. | Scales maintainable eval infrastructure and calibration standards across multiple product teams. | Designs cross-org risk tiers, authority standards and incident practices adapted to real harm. |
Map your profile and team coverage
For each of the ten competencies, record four lines: current observed level (or not yet observed), the level your role actually needs over the next 6–12 months, one piece of concrete evidence, and one deliberate action to close any meaningful gap (either by practising or by pairing with a specialist).
Interactive profile
Map your competency profile
Choose a current level and a target for each area. Leave either blank when there is not enough evidence. Choices stay in this page and reset when you leave.
Blue solid: current. Amber dashed: target. An outline appears once all ten rows in that column are assessed. Numbers match the rows below.
Use the shape to guide development and team-coverage discussions, not as a composite score.
Use the visual shape to guide development conversations, never as an averaged total or automated promotion formula. Role shapes vary legitimately:
- Core workflow PM on an established product often needs level 3 in customer understanding, delivery and product definition, with a solid level 2 baseline in model fluency and evals.
- Agent / AI-native product PM owning autonomous customer actions needs level 3+ in evaluation, operating judgement and model fluency alongside strong customer discovery.
- Platform / AI infrastructure PM skews deeper on model harnesses, observability, security boundaries and evaluation tooling.
- Product leader (Head / VP) prioritises strategy, commercial judgement, team design and decision systems while retaining enough hands-on technical fluency to spot flimsy demos and ungrounded evals.
Run the same mapping at the team level: list every consequential activity and verify a real person owns it. A strong average across a team still fails if nobody owns pausing a broken agent or maintaining source policies.
Plan your career as PM work shifts
Product management is not disappearing, yet specific flavours of it are compressing hard. Pure coordination work, writing verbose specs from second-hand notes, grooming bloated backlogs and acting as a human relay between business and engineering are losing value fast because smaller, higher-context teams and AI tools absorb that overhead.
Four distinct paths are gaining durable leverage instead:
- Hands-on AI product builder (IC): Close to users, prototypes, traces, evals and rapid shipping loops in lean teams.
- Deep domain & workflow specialist: Combines product craft with scarce vertical mastery (finance, healthcare, legal, industrial ops, regulated enterprise) where knowing the exceptions matters more than knowing the latest prompt trick.
- AI systems & platform PM: Owns shared harnesses, data/context infrastructure, evaluation platforms, pricing economics and governance primitives across multiple squads.
- Commercial product leader: Turns messy AI capabilities into viable unit economics, defensible distribution, credible enterprise trust and high-leverage AI-native teams.
Audit your week and build real proof
Look honestly at last week's calendar. Separate tasks into three buckets: repetitive formatting/coordination you should streamline, decisions you currently approve without being able to inspect the underlying mechanism, and high-judgement customer or commercial work you need more of. Pick one gap to close through real work this quarter.
Build a private, redacted portfolio of real decisions rather than chasing vanity side projects. Document the problem, constraints, alternatives you rejected, eval or commercial evidence, what shipped and what broke afterwards. Working prototypes help show initiative, while a crisp decision record, a calibrated eval rubric or a real operational turnaround carries equal or greater weight for senior roles.
When evaluating a career move, ignore AI hype in the job description and ask four blunt questions: What real customer outcome will I own? How close will I sit to real users and production data? Who will challenge my blind spots? And does leadership understand the gap between a demo and an operated product?
Try it: build a 30-day practice plan
Pick one competency where your current or target role demands more depth than your observed evidence shows, then design a concrete workplace rep rather than another passive course.
Help me design a focused 30-day workplace practice plan for one product competency.
My current role and product: [describe your workflow and scope]
Target competency to deepen: [pick one from the ten competencies]
Current observed evidence: [list real examples or "not yet observed"]
Constraints and available partners: [who can give feedback, data/access limits]
Propose:
1. One real, bounded task I can execute inside my current work over 30 days.
2. What I should do myself versus where to pair with an engineer, designer or domain expert.
3. The concrete artefact or evidence that will prove real progress.
4. One smaller alternative if I lack direct production access.
Keep the plan practical and grounded in observable work. Do not infer my level from my job title.Agree on the plan with a colleague who can give honest critique before starting. After 30 days, write down what surprised you, where you still needed help and what you would change next time. Real fluency comes from shipping and inspecting work under real constraints.