# Phase 2 → 3 Self-Assessment

> **From The Agentic TPM, Chapter 3.** Are you AI-Augmented (Phase 2) or AI-Orchestrating (Phase 3)? Most TPMs think they're in Phase 3. Most are in Phase 2. Here's the honest diagnostic.

---

## How to use this

Rate yourself (or your team) on each dimension. Be honest — the point is to know where you actually are, not where you'd like to be.

**Scoring:** 1 (early Phase 1) to 5 (deep Phase 4). Phase boundaries roughly at 1.5, 2.5, 3.5, 4.5.

---

## Assessment

### Dimension 1: How you use AI in your daily work

- **Score 1:** I've heard of AI tools. I don't use them daily.
- **Score 2:** I use AI to draft artifacts (status updates, PRDs, meeting notes). Prompts are ad-hoc.
- **Score 3:** I have a prompt library. AI is integrated into recurring workflows. I run multi-step agent chains.
- **Score 4:** I've built my own AI tools (like an AI Chief of Staff). I design workflows around agents.
- **Score 5:** I architect operating models where AI systems coordinate autonomously.

**Your score:** _____

### Dimension 2: How your team makes decisions

- **Score 1:** All decisions require human review at every step.
- **Score 2:** Some routine decisions delegated to AI drafts, human approves.
- **Score 3:** Bounded autonomy for defined decision types. Named humans Accountable per workflow.
- **Score 4:** Agent RACI in place. Meta-agent QA layer catches drift.
- **Score 5:** Autonomous decision-making with continuous supervision and human escalation.

**Your score:** _____

### Dimension 3: How you govern quality

- **Score 1:** Traditional code review, manual QA, release-gate testing.
- **Score 2:** Some automation on top of traditional QA. AI-assisted testing.
- **Score 3:** Three-tier eval harness in place. Sharpened DoD/DoR for AI work. Four-pillar gauntlet running inline.
- **Score 4:** Meta-agent validators run in production. Auditor agents catch drift and cascading failures.
- **Score 5:** Full autonomous supervision. Human review is exceptional, not routine.

**Your score:** _____

### Dimension 4: How you handle compute

- **Score 1:** Compute is IT's problem. Not something I track.
- **Score 2:** I know we have token budgets. I don't manage them directly.
- **Score 3:** I use token telemetry to prioritize work. I understand where compute goes.
- **Score 4:** Autonomous financial guardrails are armed. Per-agent quotas. Circuit breakers.
- **Score 5:** I treat compute allocation as capital allocation. Portfolio-level compute strategy.

**Your score:** _____

### Dimension 5: How your metrics work

- **Score 1:** Same metrics as before AI. Lines of code, velocity, feature count.
- **Score 2:** Some new metrics added on top of old ones.
- **Score 3:** I've dropped metrics that don't survive AI acceleration. I track outcome integrity, not activity.
- **Score 4:** Six-metric framework in production. Decision latency, context recovery cost, validation throughput integrity, etc.
- **Score 5:** Meta-metrics operational. I know how healthy my metrics themselves are.

**Your score:** _____

---

## Total and interpretation

**Total score: _____ / 25**

| Range | Phase | Meaning |
|---|---|---|
| 5–10 | Phase 1 (AI-Aware) | Observing at the edge. Time to start integrating. |
| 11–15 | Phase 2 (AI-Augmented) | Using AI as an accelerator. Next step: orchestration. |
| 16–20 | Phase 3 (AI-Orchestrating) | Designing multi-agent workflows. Redesigning coordination. |
| 21–25 | Phase 4 (AI-Leading) | Architecting operating models for autonomous systems. |

---

## What to build next

If you scored below 3 on any dimension:

- **Dimension 1 low?** Start with a prompt library. Systematize what you use.
- **Dimension 2 low?** Write your first Agent RACI for one workflow.
- **Dimension 3 low?** Build a tier-1 eval harness for one AI-generated artifact type.
- **Dimension 4 low?** Get a compute dashboard. Know where tokens go.
- **Dimension 5 low?** Pick one metric that's misleading you and replace it.

---

## Sign-off

- **Assessed by:**
- **Date:**
- **Retake in:** 90 days