Purpose
Assess whether your product work is "AI-first" (using AI to automate existing tasks faster) or "AI-shaped" (fundamentally redesigning how product teams operate around AI capabilities). Use this to evaluate your readiness across 5 essential PM competencies for 2026, identify gaps, and get concrete recommendations on which capability to build first.
Key Distinction: AI-first is cute (using Copilot to write PRDs faster). AI-shaped is survival (building a durable "reality layer" that both humans and AI trust, orchestrating AI workflows, compressing learning cycles).
This is not about AI tools—it's about organizational redesign around AI as co-intelligence. The interactive skill guides you through a maturity assessment, then recommends your next move.
Key Concepts
AI-First vs. AI-Shaped
| Dimension |
AI-First (Cute) |
AI-Shaped (Survival) |
| Mindset |
Automate existing tasks |
Redesign how work gets done |
| Goal |
Speed up artifact creation |
Compress learning cycles |
| AI Role |
Task assistant |
Strategic co-intelligence |
| Advantage |
Temporary efficiency gains |
Defensible competitive moat |
| Example |
"Copilot writes PRDs 2x faster" |
"AI agent validates hypotheses in 48 hours instead of 3 weeks" |
Critical Insight: If a competitor can replicate your AI usage by throwing bodies at it, it's not differentiation—it's just efficiency (which becomes table stakes within months).
The 5 Essential PM Competencies (2026)
These competencies define AI-shaped product work. You'll assess your maturity on each.
1. Context Design
Building a durable "reality layer" that both humans and AI can trust—treating AI attention as a scarce resource and allocating it deliberately.
What it includes:
- Documenting what's true vs. assumed
- Immutable constraints (technical, regulatory, strategic)
- Operational glossary (shared definitions)
- Evidence standards (what counts as validation)
- Context boundaries (what to persist vs. retrieve)
- Memory architecture (short-term conversational + long-term persistent)
- Retrieval strategies (semantic search, contextual retrieval)
Key Principle: "If you can't point to evidence, constraints, and definitions, you don't have context. You have vibes."
Critical Distinction: Context Stuffing vs. Context Engineering
- Context Stuffing (AI-first): Jamming volume without intent ("paste entire PRD")
- Context Engineering (AI-shaped): Shaping structure for attention (bounded domains, retrieve with intent)
The 5 Diagnostic Questions:
- What specific decision does this support?
- Can retrieval replace persistence?
- Who owns the context boundary?
- What fails if we exclude this?
- Are we fixing structure or avoiding it?
AI-first version: Pasting PRDs into ChatGPT; no context boundaries; "more is better" mentality
AI-shaped version: CLAUDE.md files, evidence databases, constraint registries AI agents reference; two-layer memory architecture; Research→Plan→Reset→Implement cycle to prevent context rot
Deep Dive: See context-engineering-advisor for detailed guidance on diagnosing context stuffing and implementing memory architecture.
2. Agent Orchestration
Creating repeatable, traceable AI workflows (not one-off prompts).
What it includes:
- Defined workflow loops: research → synthesis → critique → decision → log rationale
- Each step shows its work (traceable reasoning)
- Workflows run consistently (same inputs = predictable process)
- Version-controlled prompts and agents
Key Principle: One-off prompts are tactical. Orchestrated workflows are strategic.
AI-first version: "Ask ChatGPT to analyze this user feedback"
AI-shaped version: Automated workflow that ingests feedback, tags themes, generates hypotheses, flags contradictions, logs decisions
3. Outcome Acceleration
Using AI to compress learning cycles (not just speed up tasks).
What it includes:
- Eliminate validation lag (PoL probes run in days, not weeks)
- Remove approval delays (AI pre-validates against constraints)
- Cut meeting overhead (async AI synthesis replaces status meetings)
Key Principle: Do less, purposefully. AI removes bottlenecks, not generates more work.
AI-first version: "AI writes user stories faster"
AI-shaped version: "AI runs feasibility checks overnight, eliminating 2 weeks of technical discovery"
4. Team-AI Facilitation
Redesigning team systems so AI operates as co-intelligence, not an accountability shield.
What it includes:
- Review norms (who checks AI outputs, when, how)
- Evidence standards (AI must cite sources, not hallucinate)
- Decision authority (AI recommends, humans decide—clear boundaries)
- Psychological safety (team can challenge AI without feeling "dumb")
Key Principle: AI amplifies judgment, doesn't replace accountability.
AI-first version: "I used AI" as excuse for bad outputs
AI-shaped version: Clear review protocols; AI outputs treated as drafts requiring human validation
5. Strategic Differentiation
Moving beyond efficiency to create defensible competitive advantages.
What it includes:
- New customer capabilities (what can users do now that they couldn't before?)
- Workflow rewiring (processes competitors can't replicate without full redesign)
- Economics competitors can't match (10x cost advantage through AI)
Key Principle: "If a competitor can copy it by throwing bodies at it, it's not differentiation."
AI-first version: "We use AI to write better docs"
AI-shaped version: "We validate product hypotheses in 2 days vs. industry standard 3 weeks—ship 6x more validated features per quarter"
Anti-Patterns (What This Is NOT)
- Not about AI tools: Using Claude vs. ChatGPT doesn't matter. Redesigning workflows matters.
- Not about speed: Writing PRDs 2x faster isn't strategic if PRDs weren't the bottleneck.
- Not about automation: Automating bad processes just scales the bad.
- Not about replacing humans: AI-shaped orgs augment judgment, not eliminate it.
When to Use This Skill
✅ Use this when:
- You're using AI tools but not seeing strategic advantage
- You suspect you're "AI-first" (efficiency) but want to be "AI-shaped" (transformation)
- You need to prioritize which AI capability to build next
- Leadership asks "How are we using AI?" and you're not sure how to answer strategically
- You want to assess team readiness for AI-powered product work
❌ Don't use this when:
- You haven't started using AI at all (start with basic tools first)
- You're looking for tool recommendations (this is about organizational design, not tooling)
- You need tactical "how to write a prompt" guidance (use skills for that)
Facilitation Source of Truth
Use workshop-facilitation as the default interaction protocol for this skill.
It defines:
- session heads-up + entry mode (Guided, Context dump, Best guess)
- one-question turns with plain-language prompts
- progress labels (for example, Context Qx/8 and Scoring Qx/5)
- interruption handling and pause/resume behavior
- numbered recommendations at decision points
- quick-select numbered response options for regular questions (include
Other (specify) when useful)
This file defines the domain-specific assessment content. If there is a conflict, follow this file's domain logic.
Application
This interactive skill uses adaptive questioning to assess your maturity across 5 competencies, then recommends which to prioritize.
Facilitation Protocol (Mandatory)
- Ask exactly one question per turn.
- Wait for the user's answer before asking the next question.
- Use plain-language questions (no shorthand labels as the primary question). If needed, include an example response format.
- Show progress on every turn using user-facing labels:
Context Qx/8 during context gatheringScoring Qx/5 during maturity scoring- Include "questions remaining" when practical.
- Do not use internal phase labels (like "Step 0") in user-facing prompts unless the user asks for internal structure details.
- For maturity scoring questions, present concise 1-4 choices first; share full rubric details only if requested.
- For context questions, offer concise numbered quick-select options when practical, plus
Other (specify) for open-ended answers. Accept multi-select replies like 1,3 or 1 and 3.
- Give numbered recommendations only at decision points, not after every answer.
- Decision points include:
- After the full context summary
- After the 5-dimension maturity profile
- During priority selection and action-plan path selection
- When recommendations are shown, enumerate clearly (
1., 2., 3.) and accept selections like #1, 1, 1 and 3, 1,3, or custom text.
- If multiple options are selected, synthesize a combined path and continue.
- If custom text is provided, map it to the closest valid path and continue without forcing re-entry.
- Interruption handling is mandatory: if the user asks a meta question ("how many left?", "why this label?", "pause"), answer directly first, then restate current progress and resume with the pending question.
- If the user says to stop or pause, halt the assessment immediately and wait for explicit resume.
- If the user asks for "one question at a time," keep that mode for the rest of the session unless they explicitly opt out.
- Before any assessment question, give a short heads-up on time/length and let the user choose an entry mode.
Session Start: Heads-Up + Entry Mode (Mandatory)
Agent opening prompt (use this first):
"Quick heads-up before we start: this usually takes about 7-10 minutes and up to 13 questions total (8 context + 5 scoring).
How do you want to do this?
- Guided mode: I’ll ask one question at a time.
- Context dump: you paste what you already know, and I’ll skip anything redundant.
- Best guess mode: I’ll make reasonable assumptions where details are missing, label them, and keep moving."
Accept selections as #1, 1, 1 and 3, 1,3, or custom text.
Mode behavior:
- If Guided mode: Run Step 0 as written, then scoring.
- If Context dump: Ask for pasted context once, summarize it, identify gaps, and:
- Skip any context questions already answered.
- Ask only the minimum missing context needed (0-2 clarifying questions).
- Move to scoring as soon as context is sufficient.
- If Best guess mode: Ask for the smallest viable starting input (role/team + primary goal), then:
- Infer missing details using reasonable defaults.
- Label each inferred item as
Assumption.
- Include confidence tags (
High, Medium, Low) for each assumption.
- Continue without blocking on unknowns.
At the final summary, include an Assumptions to Validate section when context dump or best guess mode was used.
Step 0: Gather Context
Agent asks:
Collect context using this exact sequence, one question at a time:
- "Which AI tools are you using today?"
- "How does your team usually use AI today: one-off prompts, reusable templates, or multi-step workflows?"
- "Who uses AI consistently today: just you, PMs, or cross-functional teams?"
- "About how many PMs, engineers, and designers are on your team?"
- "What stage are you in: startup, growth, or enterprise?"
- "How are decisions made: centralized, distributed, or consensus-driven?"
- "What competitive advantage are you trying to build with AI?"
- "What's the biggest bottleneck slowing learning and iteration today?"
After question 8, summarize back in 4 lines:
- Current AI usage pattern
- Team context
- Strategic intent
- Primary bottleneck
Step 1: Context Design Maturity
Agent asks:
Let's assess your Context Design capability—how well you've built a "reality layer" that both humans and AI can trust, and whether you're doing context stuffing (volume without intent) or context engineering (structure for attention).
Which statement best describes your current state?
-
Level 1 (AI-First / Context Stuffing): "I paste entire documents into ChatGPT every time I need something. No shared knowledge base. No context boundaries."
- Reality: One-off prompting with no durability; "more is better" mentality
- Problem: AI has no memory; you repeat yourself constantly; context stuffing degrades attention
- Context Engineering Gap: No answers to the 5 diagnostic questions; persisting everything "just in case"
-
Level 2 (Emerging / Early Structure): "We have some docs (PRDs, strategy memos), but they're scattered. No consistent format. Starting to notice context stuffing issues (vague responses, normalized retries)."
- Reality: Context exists but isn't structured for AI consumption; no retrieval strategy
- Problem: AI can't reliably find or trust information; mixing always-needed with episodic context
- Context Engineering Gap: No context boundary owner; no distinction between persist vs. retrieve
-
Level 3 (Transitioning / Context Engineering Emerging): "We've started using CLAUDE.md files and project instructions. Constraints registry exists. We're identifying what to persist vs. retrieve. Experimenting with Research→Plan→Reset→Implement cycle."
- Reality: Structured context emerging, but not comprehensive; context boundaries defined but not fully enforced
- Problem: Coverage is patchy; some areas well-documented, others vibe-driven; inconsistent retrieval practices
- Context Engineering Progress: Can answer 3-4 of the 5 diagnostic questions; context boundary owner assigned; starting to use two-layer memory
-
Level 4 (AI-Shaped / Context Engineering Mastery): "We maintain a durable reality layer: constraints registry (20+ entries), evidence database, operational glossary (30+ terms). Two-layer memory architecture (short-term conversational + long-term persistent via vector DB). Context boundaries defined and owned. AI agents reference these automatically. We use Research→Plan→Reset→Implement to prevent context rot."
- Reality: Comprehensive, version-controlled context both humans and AI trust; retrieval with intent (not completeness)
- Outcome: AI operates with high confidence; reduces hallucination and rework; token usage optimized; no context stuffing
- Context Engineering Mastery: Can answer all 5 diagnostic questions; context boundary audited quarterly; quantitative efficiency tracking: (Accuracy × Coherence) / (Tokens × Latency)
Select your level: [1, 2, 3, or 4]
Note: If you selected Level 1-2 and struggle with context stuffing, consider using context-engineering-advisor to diagnose and fix Context Hoarding Disorder before proceeding.
User response: [Selection]
Agent records: Context Design maturity = [Level X]
Step 2: Agent Orchestration Maturity
Agent asks:
Now let's assess Agent Orchestration—whether you have repeatable AI workflows or just one-off prompts.
Which statement best describes your current state?
-
Level 1 (AI-First): "I type prompts into ChatGPT as needed. No saved workflows or templates."
- Reality: Tactical, ad-hoc usage
- Problem: Inconsistent results; can't scale or audit
-
Level 2 (Emerging): "I have a few saved prompts I reuse. Maybe some custom GPTs or Claude Projects."
- Reality: Repeatable prompts, but not full workflows
- Problem: Each step is manual; no orchestration
-
Level 3 (Transitioning): "We've built some multi-step workflows (research → synthesis → critique). Tracked in tools like Notion or Linear."
- Reality: Workflows exist but require manual handoffs
- Problem: Still human-in-the-loop for every step; not fully automated
-
Level 4 (AI-Shaped): "We have orchestrated AI workflows that run autonomously: research → synthesis → critique → decision → log rationale. Each step is traceable and version-controlled."
- Reality: Workflows run consistently; show their work at each step
- Outcome: Reliable, auditable, scalable AI processes
Select your level: [1, 2, 3, or 4]
User response: [Selection]
Agent records: Agent Orchestration maturity = [Level X]
Step 3: Outcome Acceleration Maturity
Agent asks:
Next, Outcome Acceleration—are you using AI to compress learning cycles, or just speed up tasks?
Which statement best describes your current state?
-
Level 1 (AI-First): "AI helps me write docs faster (PRDs, user stories). Saves me a few hours per week."
- Reality: Efficiency gains on artifact creation
- Problem: Docs weren't the bottleneck; learning cycles unchanged
-
Level 2 (Emerging): "AI helps with research and synthesis (summarize user feedback, analyze competitors). Saves research time."
- Reality: Modest learning acceleration
- Problem: Still sequential; AI doesn't eliminate validation lag
-
Level 3 (Transitioning): "We use AI to run experiments faster (PoL probes, feasibility checks). Cut validation time from weeks to days."
- Reality: Learning cycles compressing
- Problem: Not yet systematic; only applied to some experiments
-
Level 4 (AI-Shaped): "AI systematically removes bottlenecks: overnight feasibility checks, async synthesis replaces meetings, automated validation against constraints. Learning cycles 5-10x faster."
- Reality: Fundamental redesign of how learning happens
- Outcome: Ship validated features 6x faster than competitors
Select your level: [1, 2, 3, or 4]
User response: [Selection]
Agent records: Outcome Acceleration maturity = [Level X]
Step 4: Team-AI Facilitation Maturity
Agent asks:
Now assess Team-AI Facilitation—how well you've redesigned team systems for AI as co-intelligence.
Which statement best describes your current state?
-
Level 1 (AI-First): "I use AI privately. Team doesn't know or doesn't use it. No shared norms."
- Reality: Individual tool usage, no team integration
- Problem: Inconsistent quality; no accountability for AI outputs
-
Level 2 (Emerging): "Team uses AI, but no formal review process. 'I used AI' mentioned casually."
- Reality: Awareness but no structure
- Problem: AI outputs treated as final; errors slip through
-
Level 3 (Transitioning): "We have review norms emerging (AI outputs are drafts, not finals). Evidence standards discussed but not codified."
- Reality: Cultural shift underway
- Problem: Norms are informal; not everyone follows them
-
Level 4 (AI-Shaped): "Clear protocols: AI outputs require human validation, evidence standards codified, decision authority explicit (AI recommends, humans decide). Team treats AI as co-intelligence."
- Reality: AI integrated into team operating system
- Outcome: High-quality outputs; psychological safety maintained
Select your level: [1, 2, 3, or 4]
User response: [Selection]
Agent records: Team-AI Facilitation maturity = [Level X]
Step 5: Strategic Differentiation Maturity
Agent asks:
Finally, Strategic Differentiation—are you creating defensible competitive advantages, or just efficiency gains?
Which statement best describes your current state?
-
Level 1 (AI-First): "We use AI to work faster (write better docs,
