Run Product Like a Research Lab: Every's Dan Shipper on Building on the AI Frontier
Key points
Key takeaways from the Lenny's Podcast talk by Every co-founder and CEO Dan Shipper, recorded live at the Lenny and Friends Summit in San Francisco (September 2026):
Run your product like a research lab. Exploration of the frontier (divergent)fand execution of the roadmap (convergent) pull in opposite directions, so Shipper argues for separating concerns: a small labs team explores what new models make possible while the product team improves and scales what already works. AI makes a labs team of one affordable — one person can explore the frontier and report back what they learned.
Harness early adopters without distracting everyone. Every organisation has people who explore far more than others — they are the early adopters living in the future,but they can be a massive distraction. A labs team channels their energy and gives the rest of the product team permission to stay focused on delivery.
Two-slice teams: pirates and architects. In the AI age team size shrinks from Bezos's two-pizza teamaito one or two people. The ideal pairing: a pirate who churns out messy experiments obsessed with finding value,and an architect who shapes a chaotic prototype into something valuable, beautiful and extensible.
Make the feedback loop as tight as possible. Dogfooding is the tightest loop — build something for yourself first, otherwise test with a couple of early-adopter customers. Experiments should do real work so you can tell what is useful from what is merely new.
Run many experiments in parallel — even competing approaches. When capabilities move,the frontier becomes unknown; trying different takes on the same problem helps map it. Make experiments net-positive:turn them into external content,feed an early-adopter program and share capability learnings with the product team.
A research pipeline moves ideas from lab to product. Ideas start lab-only,prove out in real work,get adopted internally,reach early customers,and then hand off to the product team. Every reviews the pipeline weekly on its all-hands with a Notion tracker,and applies clear decision criteria — internal use as a value proxy,10x-better-than-existing,and affordability.
Case study: the copy-edit agent. Shipper spent three years automating editor-in-chief Kate's copy edits with Fable; an Every agent now runs a "Kate pass" filing suggested changes based on her historical edits. Architect Yannik turned it into a dashboard showing accepted suggestions and remaining work — Kate now does 12% less work on those edits than the month before,and it is moving toward early customers.
Labs can build the next version while scaling the current one. OpenAI's Codex was built by a small team outside the main app experimenting on form factors; its desktop app launched inFebruary 2026,grew fast enough that it merged into ChatGPT and became the foundation of that roughly 800M daily-active-user product. The sign the approach works:you welcome new model drops instead of dreading them.
Read more: Lenny's Podcast · Every
