Ai Tools Every Pm Should Use

The Core Stack (These Are Non-Negotiable)

1. Claude or ChatGPT Pro — Your Daily Partner

What it is: The foundation model you use for everything.

Why it matters: If you're an AI PM and you're not using AI heavily in your actual work, you're playing tourist in your own field. You need a daily AI practice — not to be trendy, but to build intuition about what AI can and can't do.

How I actually use it:

• Prototype features before involving engineering. I'll write detailed prompts that simulate what the AI feature should do and see how the model handles it. This tells me what's viable before we write any code.

• Draft specs and docs with AI as a thinking partner. Not "write my PRD" but "help me think through the edge cases" or "poke holes in this approach."

• Analyze user feedback at scale. Dump a batch of user feedback and ask: "What are the top 5 themes here? What surprised you?"

• Prep for stakeholder conversations by having the AI play the stakeholder. "You're a skeptical VP of Engineering. I'm going to pitch this AI feature. Push back on everything."

Which one?

Claude (Sonnet or Opus) is my go-to for most tasks. Better at nuanced analysis and following complex instructions. ChatGPT for when I need web browsing, image generation, or code execution in the same session.

Cost: ~$20-$25/month. Stop pretending this is a barrier.

---

2. Notion AI or Your Docs Tool's AI — Where Work Happens

What it is: AI integrated into your existing workflow tools.

Why it matters: The best AI is the AI you actually use. Having Claude open in a separate tab is friction. Having AI in the document you're already writing is flow.

How I actually use it:

• Summarize long threads before meetings. "Turn this 47-message Slack thread into the 5 key decisions made."

• Generate first drafts of repetitive docs. Meeting notes templates, status update structures, interview guides.

• Create tables and comparisons from prose. Paste rambling notes, get organized structured content.

• Action item extraction. Drop meeting notes, get clear next steps with owners.

What NOT to do: Don't use it for the creative thinking part. Use it for the grunt work part.

---

3. An Eval Tool — Braintrust, Promptfoo, or DIY

What it is: How you measure whether your AI features are actually good.

Why it matters: This is the defining skill gap between AI PMs and PMs who work on AI. If you can't eval, you're flying blind. "Users seem to like it" is not a metric.

How I actually use it:

• Create test sets for AI features I'm working on. 50-100 examples covering happy path, edge cases, and adversarial inputs.

• Run evals before shipping changes to prompts or models. "Did the new prompt make things better or worse?" You should be able to answer this quantitatively.

• Track quality over time. AI features drift. You need to catch it.

Tool recommendations:

• Braintrust — best for production AI teams, great UX, expensive
• Promptfoo — open source, more technical, very powerful
• LangSmith — good if you're in the LangChain ecosystem
• DIY spreadsheet — fine for starting out, doesn't scale

Start with: Promptfoo if you're technical, a simple spreadsheet if you're not. Upgrade when your eval needs get more complex.

---

4. A Prototyping Tool — Replit, v0, Cursor

What it is: A way to build AI-powered prototypes without being a full-stack engineer.

Why it matters: The best AI PMs I know can go from idea to working prototype in hours, not weeks. Not production-ready, but good enough to test and demonstrate.

How I actually use it:

• Build interactive prototypes of AI features before PRDs are written. This is game-changing for stakeholder buy-in.

• Test integrations and workflows. "What would it actually feel like to use this?"

• Create demos for customer conversations. Nothing sells like a working demo, even a hacky one.

Tool recommendations:

• v0 by Vercel — best for UI/frontend prototypes, AI generates React code
• Replit — full environment, great for quick backends and APIs
• Cursor — AI-powered IDE, best if you're comfortable with code

Start with: v0 for UI prototypes, Replit for anything with a backend. Even if you don't know code, these tools get you 80% there.

---

5. A Vector Database / RAG Setup — For Advanced Features

What it is: How you build AI features that know about your company's specific data.

Why it matters: Off-the-shelf models don't know your product, your docs, your users. RAG (Retrieval-Augmented Generation) is how you bridge that gap.

How I actually use it:

• Prototype internal tools that search company knowledge bases
• Test customer-facing AI features that need to reference product docs
• Understand the limits of RAG before promising stakeholders magic

Tool recommendations:

• Pinecone — most popular, good free tier
• Weaviate — open source, self-hostable
• Supabase pgvector — if you're already in Supabase ecosystem

Start with: You don't need this immediately. Get comfortable with the basics first. Add RAG knowledge when you're ready for advanced prototyping.

---

The Secondary Stack (Use When Needed)

These are useful but not daily drivers.

Web Research: Perplexity Pro

When you need to research a topic with AI-powered synthesis. Way better than ChatGPT's web search for actual research. I use it for competitive analysis and market research.

Transcription: Otter or Granola

Meeting transcription with AI summaries. Otter is the standard. Granola is newer and specifically designed for meeting notes + action items. Either works.

Diagramming: Whimsical AI or Miro AI

When you need to generate flowcharts, diagrams, or visual explanations quickly. The AI features are assistive, not transformative, but they save time.

Image Generation: Midjourney or DALL-E

For mockups, placeholder images, concept visualization. Midjourney produces better images. DALL-E is more convenient (built into ChatGPT). I use Midjourney for anything customer-facing.

---

Tools I've Tried and Don't Recommend

I'm going to get some hate for this, but:

ChatPRD — The prompts are too generic. You'll write better PRDs by having a real conversation with Claude and iterating. The structure isn't the hard part.

Most "AI PM Assistants" — They're wrappers around GPT with PM-specific prompts. Just learn to prompt yourself. The skill is worth more than the tool.

Any tool that claims to "automate PM work" — PM work isn't a series of tasks to automate. It's judgment, context, and relationships. Tools that miss this are useless.

47-feature enterprise platforms — You don't need an "AI-powered product management platform with integrated roadmapping and stakeholder communication and OKR tracking and..." You need Claude and a brain.

---

How to Think About AI Tools

Here's the mental model:

Tools that make you better: Use them daily. They augment your thinking, speed up your work, help you build.

Tools that replace thinking: Avoid them. If you're outsourcing judgment to AI, you're commoditizing yourself.

The best AI PMs I know use surprisingly simple tools. They're just very good at using them.

---

My Actual Workflow

Here's what a day looks like with this stack:

Morning: Check Slack. Use Notion AI to summarize anything I missed. Open Claude to prep talking points for my first meeting.

Mid-day: Working on a new AI feature. Open Replit to prototype the core flow. Run it through Promptfoo to see how it handles edge cases.

Afternoon: Writing a spec. Use Claude as a thought partner — not to write for me, but to challenge my thinking. "What would a skeptical ML engineer say about this approach?"

End of day: Quick competitive research on Perplexity. Update my eval suite with new test cases based on today's learnings.

Total AI tool cost: ~$50/month. ROI: 10x at minimum.

---

Where to Start

If you're just getting into this:

Week 1: Get Claude Pro or ChatGPT Plus. Use it for everything. Don't just ask questions — use it to build, analyze, and think.

Week 2: Add Notion AI or your equivalent. Start summarizing, organizing, and generating docs with AI assistance.

Week 3: Try prototyping something in v0 or Replit. Doesn't have to be production-ready. Just prove you can build.

Week 4: Create your first eval. Even if it's just a spreadsheet with 20 test cases. This is the skill that separates AI PMs from everyone else.

---

The Tool Isn't the Point

The PMs who succeed with AI aren't tool experts. They're experts at using AI to think better, build faster, and make better decisions.

The tools just make that easier.

Master the fundamentals. The tools follow.

---

Key Takeaways

1. You need ~5 tools, not 47. Claude, your docs tool's AI, an eval tool, a prototyping tool, and maybe a vector DB.

2. Tools that augment thinking > tools that replace thinking. Stay in the loop.

3. The skill of using AI well is more valuable than any specific tool. Learn to prompt, eval, and build.