DEVANCE. AI research lab focused on how AI changes the way we build digital products
Jul 2026 - present · devance.co · Director (part-time)
---
Devance Lab is an AI research lab about how AI changes the way we build digital products. We study what's working, what's failing, and what the frontier looks like, then help teams put it into practice. It's a collaborative, non-profit lab that works in three modes: research in focused chapters, hands-on engagements with companies, and a community of founders, researchers, and operators trading what actually works.
---
INSIGHT
Most talk about AI is about tools, models, and implementation. That's the easy part. The hard part is knowing what's worth building for a specific company, because context decides everything: data maturity, business model, operational readiness, team capability, industry dynamics. They sort real opportunities from noise. Projects rarely fail because the technology is behind. They fail because they were never fitted to a real problem, and adopting new workflows is cumbersome.
RESEARCH CHAPTERS
Who decides. How AI shifts context, access, and leverage in product decisions.
Restructuring workflow. What it takes to actually capture AI's workflow gains.
Failure modes. Why AI projects don't pay off: what breaks, at what stage, and why.
HOW IT WORKS
The lab runs on three activities. Research: chapters, each taking on one question, run with collaborators. Engagement: I approach organizations, public and private, investigate alongside them, and help integrate AI into their workforce. Community: bringing founders, researchers, and operators together to trade ideas through meetups and shared writing. Underneath sits the operating model I want to test: sign-offs from user and market teams feed a per-company context engine and AI tooling, so decisions get better and implementation gets faster.
DECISIONS
The frontier of use, not of capability. Most labs work the technical frontier: what the models can newly do. We work the other frontier: how real companies actually put AI to work in how they build. The cost is that it reads as less novel than model research, and the technical crowd may not call it research at all. The bet is that the unsolved problem isn't capability, it's fit and adoption, and that's where value actually leaks out today.
A non-profit, self-managed lab, not a company. I set this up as collaborative and non-profit rather than a startup with equity and a burn rate. Collaborators keep their full-time jobs and contribute part-time; nobody is paid in the short term. The tradeoff is real: no capital, slower pace, and I can't demand anyone's time. What it buys is freedom to pick honest questions, and a low bar for smart people to join for the credibility and the peers rather than a paycheck.
Chapters, opened by curated contributors. Instead of a fixed agenda, the unit is a chapter: one question, owned by a contributor who can also bring in an engagement. But it's curated, not open to any topic or anyone: I choose the people and the questions, so the work stays coherent and sharp. It scales through people I don't have to manage, at the cost of pace, since curation is deliberately a bottleneck. I'd rather trade volume for range and real-world access at a quality bar I trust.
STATUS
Just started, in July 2026, and run part-time. Next steps: gather the first collaborators, and start talking to practitioners to understand how people are actually using AI in their work.