That's the fear. You've seen the headlines. 62% of tech layoffs cite AI. Your VP asked if the team is "AI-first." You tried ChatGPT — it wrote something generic. You're not sure whether to be more afraid of AI or more afraid of being left behind by PMs who figured it out.

Here's what the data actually says: PMs spend 43% of their working week on documentation, meetings, and status updates. That's roughly 17 hours of work that requires no human judgment — conversion work. Reformatting. Summarizing. Repeating the same information to different people.

AI is not coming for your job. It's coming for the 17 hours. And you actually want that.

Primary watch: Claude for PMs (Cowork + Code + Dispatch) by Aakash Gupta (growproduct)
Link:

https://www.youtube.com/watch?v=bITUsUsrxjM

Duration: ~60 minutes (focus on chapters covering Projects, persistent context, and PM workflow patterns)
What it teaches: The definitive Claude course for PMs — building Projects with persistent context, using Claude as a PM's documentation engine, and workflow patterns designed specifically for product managers. Aakash Gupta is one of the top PM thought leaders in AI.
What it misses (our context layer adds): The video teaches Claude for PMs broadly. We add: the five PM-specific mental models that turn Claude from a chatbot into your documentation engine, the AI-feasible vs human-essential task framework, and the reframe that makes AI feel like a capacity multiplier instead of a career threat.

Quick reference (7 min): Claude Projects Step-by-Step by Kevin Stratvert
Link:

https://www.youtube.com/watch?v=w7_yWjYyxjE

What it teaches: Pure Claude Projects focus in 7 minutes. The gold standard for accessible, no-fluff tutorials.

If you already watched the Aakash Gupta tutorial in M0-S1, focus on the prompting fundamentals and Projects sections this time. You're watching with new eyes now — you know what Claude can do because you just built something with it.

Before we get into the models, let's use AI right now — because this course is AI-first, not AI-later.

1. Open your "My Weekly Reviews" Project from S1 (or create a new chat at claude.ai)
2. List 6-8 recurring tasks you do every week — be honest, include the admin stuff:
   • e.g., "Write Monday standup update for engineering team"
   • e.g., "Review 3 competitor product changelogs"
   • e.g., "Sit in on customer discovery calls and take notes"
   • e.g., "Decide which features to prioritize for Q3"
3. Copy this prompt and paste it into Claude with your task list:

Here are my recurring work tasks. For each one, classify it as:
- AI-FEASIBLE: AI can do or significantly help (>50% time savings)
- HUMAN-ESSENTIAL: Requires my judgment, relationships, or strategic thinking

Give a one-sentence reason for each classification.

[PASTE YOUR TASKS HERE]

4. Read Claude's classifications. You will probably disagree with 1-2 of them. That's the point — the disagreement teaches you where your judgment lives.

5. Keep Claude's output open. We'll compare it to the mental models below.

Why this works: You just used AI to analyze your own work — and you disagreed with it. That friction is the entire lesson of this session. AI can process and classify. You hold the context it can't see.

These aren't abstract concepts. They're the operating system for how you'll think about AI from this point forward. Each one maps directly to something you just experienced in M0-S1.

1. AI as Junior Analyst, Not Oracle

AI excels at processing what you give it — synthesizing, reformatting, summarizing, extracting patterns. It cannot read your customers' minds. It cannot intuit your product strategy. It cannot tell you which feature to build.

What this means: You define the assignment. You provide the context. You verify the output. You don't ask AI "what should we build?" You ask it "here's all the evidence — what patterns do you see?"

You experienced this in M0-S1: You gave Claude your voice document (context) and this week's bullets (assignment). Claude assembled them into a review. It didn't decide what to include — you did. It didn't know your team dynamics — you filled those in during the 3-minute edit.

2. Garbage-In-Garbage-Out with Context

The single biggest reason PMs dismiss AI as "generic slop": they ask "write me a PRD" with zero context. Claude fills the void with clichés because that's all it has. Give it your last PRD, your customer interviews, your Jira tickets — the output transforms.

What this means: Every time you're disappointed by AI output, ask: "What context did I provide?" Not "What prompt did I write?" The context is the difference between generic and useful.

You experienced this in M0-S1: The prompt template says "using the reference document I uploaded." Without that document, Claude writes a generic status update. With it, Claude writes YOUR status update in YOUR voice.

3. The PM's Judgment Is the Moat

AI can compress 20 hours of admin per week. It cannot make trade-off decisions. It cannot sit across from a VP and say "we're deprioritizing your request because the data shows it serves 3% of users." It cannot read the room in a stakeholder meeting where politics meets product.

What this means: Everything AI writes for you needs your judgment before it ships. The 3-minute edit isn't quality assurance — it's where you add the strategic context the AI doesn't have.

You experienced this in M0-S1: You caught AI-voice words. You fixed facts. You adjusted tone. Those edits are your judgment — the thing AI can't do.

4. Quality Gates Are for AI Too

PMs apply review cycles to human work — peer review on PRDs, QA on features, legal review on terms. AI output deserves the same rigor. Before you ship anything AI wrote, ask: Does this pass my defined constraints? Does it use my domain language? Would I stake my reputation on it?

What this means: Never forward AI output directly. Always review. Build a checklist (you'll create yours in Module 4). The AI is the writer. You're the editor. Editors get the byline.

You experienced this in M0-S1: The 3-minute edit IS a quality gate. You checked facts, tone, and AI-voice. That's the minimum gate. Later modules add more.

5. Build Your System, Not Your Prompts

The difference between a PM who "uses ChatGPT" and an AI-first PM is architecture. The AI-first PM has a reusable template for every recurring task, context that persists across sessions, and output that chains from one step into the next. They're not writing new prompts every Monday — they're running their system.

What this means: Every time you find yourself writing a prompt from scratch, ask: "Should this be a template?" The answer is almost always yes. Templates compound. Systems scale. One-off prompts are a tax on your future self.

You experienced this in M0-S1: You built a Claude Project with persistent context. Next Monday, you won't write a new prompt. You'll open the Project, paste bullets, and generate. That's a system.

Now let's get specific about YOUR work. Here are the most common PM tasks, classified by whether AI can handle them or you need to:

AI-Feasible (AI Can Do or Help Significantly)

Human-Essential (Requires Your Judgment — AI Cannot Do)

Here's what changes after this session:

Before: "AI is going to automate my job."
After: "AI can automate 43% of my week. I get to spend that time on strategy, customers, and decisions — the work I actually became a PM to do."

Before: "I tried ChatGPT and it wrote generic slop."
After: "I gave ChatGPT zero context and expected it to read my mind. When I give AI my actual documents and specific bullets, the output is usable."

Before: "The engineers on my team use AI tools I don't understand."
After: "I have my own AI system — one built for PM work, not for code. It saves me hours every week. I don't need to code to benefit from AI."

Write down your own reframe. What's the one fear about AI you came in with? What's the one capability you saw today that changes how you think about it?

In M0-S3, you'll learn to evaluate ANY AI tool in 15 minutes — so you never waste time on demos again. The tool landscape changes monthly. The evaluation framework doesn't.