Delivery Isn't Free

Show Notes

• Petra Wille and Teresa Torres take on the popular claim that AI has made software delivery essentially free — and explain why that framing falls apart the moment you care about shipping something real.

• Why "building a single thing" getting cheaper doesn't mean delivery is cheaper — especially when teams respond by building far more things

• The three-week feature spiral: how "free" delivery quietly produces Frankenstein data models, duplicated code in 17 places, and tanked performance

• Why you still need a skilled engineering team observing and steering what AI produces — and why architecture decisions can't be outsourced to a coding agent

• Build to learn vs. build to earn (via SVPG): throwaway prototypes for discovery are dirt cheap and great — as long as you actually throw them away

• The app store paradox: apps released are spiking, but apps actively used stay flat — and how users are getting sharper at sensing "AI app slop"

• Why AI products are a different beast than deterministic code — error analysis, evals, LLM-as-judge, prompt and orchestration iteration all take real time

• A candid behind-the-scenes look at building AI-generated opportunity solution trees with Vistaly, and the quality issues only a domain expert can catch

• The 60–70% trap: getting to a good-looking prototype is fast; closing the last 30% to a trustworthy product is months to years of work

• The real hidden sentence: "Delivery is free" is almost always followed by "…so taste is all that matters" or "…so discovery is all that matters." Both are wrong. Delivery and discovery both matter — forever.

Notable Moments

"I don't think delivery is free. I don't think delivery will ever be free."

"By the time you're getting to feature 15, your data model looks like a Frankenstein strategy."

"The first 60 to 70% is easy. It's a prototype… Closing that last 30% is months to years of work."

Resources & Links:

• Follow Teresa Torres: ProductTalk.org

• Follow Petra Wille: Petra-Wille.com

Mentioned in this episode:

• Just Now Possible (justnowpossible.com) — Teresa's podcast on what's newly possible with AI in product

• SVPG’s Build to Learn vs Build to Earn (svpg.com/build-to-learn-vs-build-to-earn) — origin of the "build to earn vs. build to learn" framing

• Vistaly (vistaly.com) — Teresa's tool for AI-generated snapshots and opportunity solution trees (closed beta → GA)

• Opportunity Solution Trees (OSTs) (producttalk.org/opportunity-solution-trees) — Teresa's continuous discovery framework

• AI Evals & Discovery - All Things Product Podcast with Teresa Torres & Petra Wille (producttalk.org/ai-evals-discovery-all-things-…) — Past episode to read/watch/listen on about evals, error analysis & LLM-as-judge — the discipline behind building reliable AI features

• Claude Fable (anthropic.com/claude/fable)

• Facilitation Skills - All Things Product Podcast with Teresa Torres & Petra Wille (producttalk.org/facilitation-skills-all-things…) — Past episode where Teresa and Petra talked about how many apps are being added to the app store, but not being used.

• Taste - All Things Product Podcast with Teresa Torres & Petra Wille (producttalk.org/taste-all-things-product-podca…) — Teresa thinks taste is baloney, Petra and her talked about that in this episode