The most important AI governance story this week does not look like a governance story on the surface. It looks like a medical milestone. A Harvard-led trial published in the journal Science put OpenAI's o1 reasoning model in an emergency room alongside pairs of human doctors, gave them identical patient records, identical time, and identical intake information, and the AI diagnosed correctly 67% of the time against 50 to 55% for the physicians. That result will travel through every health system boardroom in the world over the next 90 days. The conversation it will generate will almost entirely miss the point. The question everyone will ask is whether AI can replace doctors. That is the wrong question. The right question is this. When the AI says one thing and the doctor says another, and the AI turns out to be right, who carries the liability for the decision the human made? And when the AI turns out to be wrong and the doctor followed it without sufficient scrutiny, who carrie...

Six weeks. Twelve posts. One argument that sharpened with every piece I wrote. The argument is this. The capability layer of enterprise AI is accelerating faster than the governance architecture in every direction simultaneously. Not in one domain. In every domain at once. Output trust, agentic infrastructure, vendor sovereignty, architectural uncertainty, budget governance, decision quality, release pipeline security, and the quiet structural bias of systems trained to agree rather than challenge. Each post below took a different entry point into that same governing reality. Read together they form a single extended argument about what responsible enterprise AI deployment actually requires in 2026 and why almost no organisation is fully ready for it. This is the map. A systems view of the governance fractures emerging across enterprise AI adoption. The Research Foundation Everything in this series rests on a research question explored in co-authored work presented at BIGS 2025. The qu...

The most dangerous AI in your organisation right now is not the one making obvious mistakes. It is the one making you feel right about the wrong things. This is the agreement loop. It is quieter than a hallucination, harder to detect than a bias error, and considerably more consequential than either. It does not announce itself. It compounds. Every validation the system offers makes the next question slightly more confident, the next assumption slightly less examined, the next decision slightly further from the scrutiny it deserves. By the time the loop becomes visible it has already shaped the decision, the strategy, or the recommendation that no one questioned because the AI kept agreeing. How the agreement loop actually works Large language models are trained on human feedback. The feedback mechanism that makes them useful, responding helpfully, adapting to the user's context, maintaining conversational coherence, is also the mechanism that makes them susceptible to what resea...

There is a pricing shift happening underneath the AI capability headlines that most enterprise teams are not tracking closely enough, and Perplexity's numbers from March 2026 just made it impossible to ignore. Annualised recurring revenue jumping from $305 million to $450 million in a single month. A 50% revenue increase in 30 days. More than 100 million monthly active users. And a business model that has quietly pivoted from being an AI-powered search engine to being a platform for building businesses, with a $1 million competition to prove it. The Perplexity story is being covered as a growth story. It is actually a pricing architecture story. And the implications for how enterprises budget, procure, and govern AI spend over the next 24 months are more significant than the headline numbers suggest. We need to spend time on both sides of this. The capability and growth story is interesting. The business model transformation underneath it is the part that will land on enterpris...

Seven models. Simultaneously. On a single supercomputer cluster. One of them with 10 trillion parameters, a scale that would make it among the largest AI systems ever trained. When Elon Musk revealed that xAI is running all of this in parallel on Colossus 2, the coverage focused almost entirely on the raw numbers. The parameter counts. The infrastructure scale. The competitive positioning against OpenAI and Anthropic. That is the wrong frame for understanding what this moment actually signals, and what it means for every enterprise team currently making AI platform decisions. The number that matters most is not 10 trillion. It is seven. Seven simultaneous training runs does not say "we have found the answer and we are scaling it." It says "we do not yet know which architectural approach will work and we are running multiple bets in parallel to find out faster." That is a fundamentally different signal from what most AI infrastructure announcements communicate, and...