Back in November 2015 I wrote that building products customers actually want is really hard, and that most product efforts fail. I still believe that. What I've changed my mind about is what I thought the fix was.

That's the year I put out the Product Discovery Canvas, drawing on people like David Hussman, Steve Blank, Jeff Patton, Alberto Savoia, and Eric Ries. The idea behind it was simple: real learning comes from real users interacting with real things, not more meetings, not better decks. I told teams blanks were okay, that an empty box was just an honest picture of what you didn't know yet. And I pushed pretotyping hard, fake it before you make it, because making it was expensive.

Eleven years and a lot of client work later, I'd still defend every bit of that thinking. What's changed is the economics it was built on. Pretotyping existed because building the real thing meant months of engineering and real budgets on the line, so faking it was the smart move. Here's the argument I want to make: agentic AI hasn't made pretotyping obsolete, it's produced a new kind of pretotype. The old one avoided building the mechanism because the mechanism was expensive, a landing page or a human behind the curtain standing in for the real thing. The new one builds the mechanism, disposably, because an agent can now do it cheaply enough to throw away. Same instinct, compress the distance between idea and evidence, running on different material.

In this talk I'll go through what held up from the canvas, what I got wrong, and what we actually do now, a 6-Week Product Sprint that runs those two lanes of discovery, the old pretotype and the new one, through weeks of building with real users instead of a wall of sticky notes. I'll get into why a failed checkpoint in week two counts as a win, and why I think discovery didn't go away, it just stopped producing documents and started producing working software.

Target Audience
Product managers, product owners, engineering leads, startup founders, and agile practitioners who've used (or fought with) discovery frameworks like the Lean Canvas, Design Sprints, or story mapping, and want to know what still holds up now that AI Agents can build.

Duration
30 minute talk, including Q&A

Learning Outcomes
•Apply the enduring discovery principles (shared understanding, honoring the unknown, validating before building) regardless of tooling.

•Tell the two lanes of pretotyping apart: the classic kind (landing pages, paper prototypes, Wizard-of-Oz tests) for questions about desire and willingness to pay, and the new kind, a disposable but real AI-built version, for questions about whether the thing actually works.

•Structure a discovery effort as a phased, gated sprint (concept to working version to beta to launch-ready) with acceptance/behavioral tests defined up front.

•Reframe a failed validation checkpoint as a successful, cheap outcome rather than a failure to be hidden.

This closes out an 11-year, multi-part series on product discovery that started with the Product Discovery Canvas in 2015. It's reflective, not a product pitch, and works as an opening or closing session or a track talk for product/agile audiences.

Companion blog post: https://agileinnov.com/blog/product-discovery-eleven-years-on-from-the-canvas-to-the-sprint