October 13, 2026 · 6:04 AM PT edition
Mind Blow: agent license plates, AlphaFold for Chagas and India's AI scale play
Wrong AI races, biology meets agents, and builders budgeting context overnight.
Only 8% of organizations have a rule for what happens to time that AI saves. Tim Berners-Lee’s fix for runaway agents is blunt: give them license plates.
Here is what happened in AI and DevOps this cycle.
John Romeo of the Oliver Wyman Forum argues that everyone is watching the wrong AI race. Labs pour hundreds of billions into chips, models and data centers for raw intelligence, but he says the decisive contests are adaptation and control. Adaptation means rewriting workflows around the model, not stacking pilots that never touch the bottom line. An insurance claim can finish in seconds and still sit three days in an approval queue. Give an agent data, a budget and permission to act, he warns, and you have created an economic actor. Identity, permission, accountability and revoke-ability are the checklist. Sixty-eight percent of employees say they would take a pay cut for a guarantee that AI would never replace them. The machines may play the game better. Humans still have to write which games they are allowed to play.
Nandan Nilekani, Infosys co-founder and architect behind Aadhaar and UPI, says India will lead where AI actually gets used. Aadhaar now covers roughly 1.3 billion people, and UPI pulled hundreds of millions into digital payments. New models land every week in India, the US and China, he notes, but diffusion at population scale is where the action shifts once the supply-side chip race cools. Coding agents will automate a lot of classical coding. That is only about 30% of any transformation. The other 70% is governance, change management and making sure agents do not go off the rails. Small businesses of three to five people, he argues, still do work that is hard to automate, and that is where India’s next jobs will pile up.
At the same Bloomberg New Economy Forum, Moderna’s Noubar Afeyan, DeepMind’s Lila Ibrahim and Biocon’s Kiran Mazumdar-Shaw mapped what AI is doing to science. Ibrahim points to AlphaFold, now as easy as a map search for millions of researchers. Working with teams in Spain, it helped reconnect an FDA-approved drug to Chagas disease, which affects about 8 million people, a link humans had not made. Afeyan is cooler on digital twins for drug testing and hotter on triage: early cancer screens can create ten times more false positives than true ones, and AI that cuts that flood would change clinical care without pretending cancer is cured next Tuesday. He also says 2026 crossed a line. LLM-spawned agents that adapt and swarm start looking like living organisms, and he wants the people who commercialize agents, not the airlines they book for, to carry the regulation.
Builders are treating context itself as infrastructure. Suraj Sharma’s first “must-build” for AI engineers is a Dynamic Context Assembler: middleware that budgets tokens across working memory, tool schemas and retrieved docs on every request. His line is that prompt engineering is dead and context overflow is the silent killer. Mori sketches the overnight version of the same idea, claiming Grok Bot plus Opus 5.5 can coordinate agents, fix GitHub issues and prep pull requests while you sleep. Treat the 24/7 engineering-team pitch as a demo claim until you see the diffs.
Andriy Burkov notes that strong results on code and multi-step reasoning usually need a mid-training phase: continued pre-training on expert data between ordinary pre-training and RL. Kushal Gangil points at a free open-source curriculum, AI Engineering from Scratch, described as 503 lessons across 20 phases and about 320 hours, aimed at people who want to build the ideas behind the frameworks rather than only call them. Divyansh’s roundup of open-source “cheat code” repos for AI builders leads with Browser Use for agents that drive a real browser. Worth a look, not a syllabus.
That’s the cycle. Put a license plate on your agents, budget their context, and see you tomorrow.
Also on the radar: these also appeared this cycle, but there was no transcript to report from yet (captions rate-limited by YouTube, no English captions, or no speech), so they aren’t summarized here: The Only Vim Tutorial You Actually Need (Vimtutor) (That DevOps Guy); Judge a Transaction by the Company It Keeps — Somya Rai & MD Sharique, EXL (AI Engineer); I Believe Everything My Agents Tell Me: Three Levels of Review — Corey Gallon, Rexmore (AI Engineer).