PDP — Product Discovery Protocol

Product Discovery Protocol — PDP for short — is a single YAML file that describes everything an AI agent needs to demo your product: the pages it can navigate to, the flows it can walk a visitor through, the demo accounts it can sign in with, and the knowledge it can answer questions from.

Once you upload a PDP for your product, the iraa.ai demo agent uses it to run live, interactive demos for every website visitor — clicking buttons, filling forms, narrating value, answering questions, and capturing leads. No pre-recorded tour. No scheduling. The agent is operating your real product.

Frontier LLMs can reason about a UI, but no model has been trained on your product. To run a real demo, the agent needs ground truth: the routes that exist, the selectors that survive a refactor, the flows you actually want a prospect to see, the FAQs you'd answer the same way every time. Stuffing all of that into a system prompt gets brittle fast.

PDP is a structured, versioned, validated answer. Authoring is the once-per-product cost; everything downstream — the demo, the agent's answers, the lead capture, the analytics — keys off of it. It's the same instinct as MCP for tool use: a small, stable contract between the model and the world.

And because it's a file you check into your repo, PDPs evolve with your product. When you ship a new module, you ship a PDP update. The demo agent stays in sync without anyone re-recording a tour.

You don't write the YAML by hand. From the iraa.ai dashboard, copy the authoring prompt and paste it into an AI coding agent (Claude Code, Cursor, Lovable, Replit). The agent reads your repo, asks you a few questions about demo accounts and the order you want flows to play in, and writes the PDP.

Validation is free — the dashboard lints the PDP against the schema before upload, so you never ship a broken demo. Updates are the same prompt, in drift-aware mode: the agent diffs your repo against the existing PDP and applies the delta.