Synthesis: AI Strategy

The Big Picture

You’ve worked through four lessons: when AI is the right solution, how to choose between Build/Buy/Bake, how AI can threaten existing product-market fit, and where the biggest AI opportunities lie.

Individually, these are decision-making tools. Together, they form a strategy logic: Lesson 1 asks IF AI is the right approach. Lesson 2 asks HOW to implement it. Lesson 3 asks WHAT’S AT STAKE. Lesson 4 asks WHERE the biggest leverage is.

The 5 check questions (L1) filter the opportunities (L4) that protect your PMF (L3) using the right implementation level (L2).

Connections

1. AI is not a feature — it’s a new paradigm

This runs through all four lessons. Reforge puts it bluntly: “Not about sprinkling AI features — fundamentally rethinking how you compete.” Elena Verna argued in a conference analysis that 60–70% of traditional growth tactics no longer work in the AI era. And the 64% user rejection rate of bolt-on AI (ZDNET) proves: tacking it on isn’t enough.

For you as a PM: AI isn’t a feature request you prioritize. It’s a strategic decision that reshapes your product, your business model, and your competitive position.

2. The speed of disruption is unprecedented

ChatGPT: 1 million users in 5 days. Cursor: $0 to a $9B valuation in roughly 18 months. Chegg: 90% PMF collapse in 9 months. The windows for reaction are dramatically shorter than in any previous technology wave.

For you as a PM: Quarterly planning isn’t enough. You need PMF monitoring (L3) and an opportunity pipeline (L4) that can reprioritize fast.

3. Winners use AI as amplifier, not replacement

Duolingo uses AI to make its core product better — result: +51% DAUs. Notion embeds AI into existing workflows — over 50% adoption. The losers? Chegg and CNET tried to replace their core value with AI. Klarna failed with the replacement approach first, only succeeding once they switched to a hybrid model.

For you as a PM: The build-vs-buy decision (L2) isn’t just technical. It determines whether AI strengthens your product or hollows it out.

4. Data is the only lasting moat — the Data Flywheel

Public data offers no defensibility — Chegg and Stack Overflow learned this the hard way. Proprietary data from user interactions does — that’s the leverage behind Duolingo and Notion. Models are commodities. Data and UX are not.

The strongest pattern across successful AI products is the Data Flywheel: More users → more interaction data → better model → better UX → more users. This cycle builds a self-reinforcing advantage that competitors cannot simply copy. Reforge argues that proprietary data is the only durable competitive moat in AI — models get better and cheaper, algorithms get copied, but your specific data flywheel belongs only to you.

For you as a PM: Your PMF defense (L3) and your implementation level (L2) must build toward proprietary data. If you’re only wrapping APIs, you have no moat. For every feature, ask: Does it generate data that makes the product better?

5. “Start with API, build the last mile”

Nearly every successful AI product starts with APIs (L2) and then differentiates through proprietary UX, data, and workflows. Cursor is the poster child: API wrapper plus excellent UX equals a $29B valuation. The differentiation isn’t in the model — it’s in the last mile.

For you as a PM: Starting with APIs (L2) isn’t a compromise — it’s the fastest route to validation. The opportunity (L4) lies in what you build around it.

The Meta-Insight

AI strategy is not a technology decision — it’s a business model decision. The question isn’t “Can we use AI?” but “Does AI change what our users pay us for — and are we on the right side of that shift?”

This is what sets AI strategy apart from every previous build-vs-buy choice. You’re not just deciding on implementation. You’re deciding whether your product will still be relevant in 12 months.

Your Strategy Checklist

What you should now be able to do:

Evaluate whether a problem needs AI or rules suffice — Lesson 1
Choose the right implementation level (prompt → API → RAG → fine-tuning) — Lesson 2
Assess your PMF risk from AI disruption — Lesson 3
Identify and prioritize AI opportunities using RICE-A — Lesson 4
Recognize and prevent vendor lock-in — Lesson 2
Spot PMF collapse signals early — Lesson 3

If any of these feel uncertain, go back to the relevant lesson. These strategic foundations determine whether your AI initiatives create value or burn resources.

Continue with: AI Product Design

You know WHEN AI makes sense. Chapter 3 shows HOW to design AI features that users trust.

Self-Assessment

Three scenarios combining multiple concepts from this chapter. Think through your answer before revealing the solution.

Scenario 1: The AI Feature Rush

Your CEO returns from a board meeting and says: “Our competitors all launched AI features. We need one by next month.” Your product is a B2B project management tool with stable PMF. How do you respond?

Solution

Start with the 5 check questions (Lesson 1): does the problem actually need AI, or would rules suffice? Then use opportunity identification (Lesson 4) to find where AI creates the most leverage — don’t just copy what competitors are doing. At the same time, assess the PMF risk (Lesson 3): your stable PMF means bolt-on AI could actually threaten it (64% user rejection rate per ZDNET). The right answer isn’t “no” but “yes, strategically” — with the right implementation level (Lesson 2), starting with an API integration and differentiating through proprietary data.

Scenario 2: The API Wrapper with an Expiration Date

Your startup built a successful AI writing tool using GPT-4 via API with a polished UX on top. You’re growing fast, but an investor asks: “What’s your moat? OpenAI could launch this themselves tomorrow.” What’s your strategy?

Solution

The investor has a valid point: pure API wrappers have no lasting moat (Connections 4 and 5). Your strategy must build toward proprietary data (Lessons 2 and 3): user interaction data, domain-specific templates, individual writing style profiles. The Cursor example shows that API-start plus excellent UX works — but only if you build the “last mile” (Connection 5). Simultaneously, avoid vendor lock-in (Lesson 2): build multi-model capability so you’re not dependent on a single provider. The moat isn’t in the model — it’s in the data layer and UX depth.

Scenario 3: PMF Under Fire

You’re a PM at a company offering online tutoring. Since ChatGPT, students can ask their questions directly to AI, and your usage numbers have been declining for three months. The CTO proposes building an AI chatbot to replace the human tutors. What do you do?

Solution

This is a classic PMF collapse scenario like Chegg (Lesson 3): your core value — answers to learning questions — is threatened by free AI alternatives. But the Chegg lesson also shows what doesn’t work: replacing your core value with AI. Winners like Duolingo use AI as an amplifier (Connection 3): AI makes human tutors better, not obsolete. Concretely, you could deploy AI to support tutors with real-time explanation suggestions, create personalized learning paths, or be available between sessions. The opportunity (Lesson 4) isn’t in replacing the human element but in amplifying it — because that’s exactly what ChatGPT alone can’t offer.

Sources: Building on Lessons 1–4. Reforge AI Strategy Framework (2024), Elena Verna Growth Analysis (2024), ZDNET/Aberdeen Consumer AI Survey (2024), Duolingo Q4 2024 Earnings, Notion AI Adoption Data (2024), Chegg SEC Filings (2023–2024), Cursor/Anysphere Financial Data (2025)