a sovereignty gap in science
AI is making scientific possibilities abundant. The next bottleneck is turning them into verified physical knowledge and reproducible industrial capability.
shyamal’s space
something new in AI x science • ex-Applied AI @ OpenAI • AI Advisor to Startups • On Deck Fellow • Proud Son • Duke + Wisconsin Alum • Building for impact • Venture Scout • Neo Mentor • Duke AI Advisory Board
Latest posts
some thoughts on a post agi world and economy
the economy is mostly atoms + energy + instructions. atoms are what you want (house, a meal, a road, a phone). energy is what moves atoms. instructions are what tell energy how to move atoms. for the last ~200 years, instructions were scarce. coordination was hard. planning was expensive. management was a tax. so we built companies to bundle cognition + capital + labor. agi flips that. when…
clarity for builders
clarity is the hidden unfair advantage. when you know exactly what you want to build, you move faster, you avoid the fake work, and you say no to every shiny distraction. most people drift; the ones who get somewhere decide where they’re going and keep checking the compass. clarity is the compass.
How to run a research loop and a product loop at your company
Most startups that try to do both research and product end up doing neither well. The research group drifts into science projects, the product group starves for breakthroughs, and everyone is frustrated. This is a classic failure mode.
On meetings
Meetings are expensive. Default to async.
Experience × Evals: the build loop for self‑improving agents
author’s note: this essay summarizes and synthesizes ideas from the linked pieces and adds my own perspective to extend them.
Age of Taste: where the cost of trying is ~zero, but knowing what to do is everything
work will re‑orient around what cannot be sped up by more compute. in other words, with general intelligence, work humans do shall migrate to the handful of domains where an extra tera FLOP confers little or no advantage. the thesis is simple : as the cost of exploring ideas approaches zero, human focus inevitably shifts to areas where raw computational power provides diminishing returns. the true…
Age of the Agent Spec
Age of the Agent Orchestrator
Over the past couple of years at OpenAI, many friends and family have often asked me what the world might look like when AI and agents can do “everything”. it’s a question we’re still collectively wrestling with. while that’s not happening anytime soon, capabilities are advancing quickly enough that it’s worth thinking a step ahead. One frame that I find useful is to ask what becomes scarce.…
Reinforcement fine-tuning and reward engineering
RFT is still in its early days, but it’s a powerful tool - making it surprisingly easy to create expert models for specific domains with less training data.