PM Product Teardown

AI-Powered Competitive Analysis Engine

“I built the tool I kept wishing existed.”

outcomes

Stack wired: Next.js 14 + Firecrawl + Anthropic SDK + Supabase
Core prompt architecture defined
Teardown output: problem, value prop, prioritization assumptions, GTM signals, gaps

pm skills

competitive researchframework application (JTBD, RICE, opportunity sizing)product strategydeveloper tooling for PMsmeta-product thinkingstructured LLM output design

Problem

PMs do product teardowns manually. Screenshot, annotate, synthesize. It's slow, inconsistent, and doesn't scale. The frameworks already exist, jobs-to-be-done, opportunity scoring, RICE, but applying them systematically to any product URL takes hours. There's no tool that does this for you.

What I Built

A web app where you input a product URL and receive a structured PM teardown.

Firecrawl handles scraping. Claude applies PM frameworks, not summarization, interpretation. Supabase stores each teardown for side-by-side comparison.

Output Field	Description
Problem	Core user pain, inferred from page copy and positioning
Value Prop	What the product promises to solve
Priorities	Feature bets implied by the UI
GTM Signals	Pricing, ICP, and positioning language
Gaps	What competitors or alternatives do better

Approach

Designed the output schema before the UI. A teardown is only useful if it's structured and comparable, so the LLM is prompted to return a consistent shape. Unstructured teardowns aren't actionable. That's the product instinct: structured output is a requirement, not a nice-to-have.

PM Skills Applied

Meta-product thinking: applying PM rigor to PM tooling. The tool makes explicit what most PMs do implicitly. That means it can be shared, iterated on, and taught, which makes it a product, not a workflow.

Technical Depth

Next.js 14 App Router, Anthropic SDK (Claude structured output), Firecrawl JS (web scraping), Supabase (teardown persistence and retrieval), shadcn/ui, TypeScript, Tailwind. Prompt architecture uses chain-of-thought elicitation followed by structured JSON extraction.

Lessons

Lesson	Why It Matters
Schema-first matters more for AI tools than CRUD apps	The prompt IS the product
Firecrawl + Claude = "read anything, analyze anything"	Works on any URL without an API
Meta-PM angle is a differentiator	Building tools for PMs is itself a PM skill

The Story

(WIP, scaffold and core architecture complete, full UI and output refinement in progress)

Every PM I know does competitive teardowns in a Notes app or a messy Notion doc. The insight was simple: the frameworks already exist, the LLM can apply them consistently, and Firecrawl means it works on any URL. The interesting part is the prompt design, getting Claude to produce a teardown that sounds like a sharp PM wrote it, not a summarizer.