The people who understand AI best are all quietly building their own personal versions. Not using the products. Building. There's something worth paying attention to in that.
Everyone's asking where the moats are when AI tooling is everywhere. One answer is becoming clear: codified organisational knowledge that can be acted upon at machine speed.
We're moving toward production code that no human ever reads. The bottleneck isn't trusting the code — it's building the human capacity to design trustworthy systems.
What's happening to listed SaaS companies is exactly what happened to translation companies in 2021. Markets don't fear AI — they fear the collapse of their predictive frameworks.
Anthropic's Claude Cowork plugins wiped tens of billions off software valuations in days. This is what disruption looks like in 2026 — a cliff edge, visible in real time.
The Vatican brought together 50 experts for a serious AI governance summit. What they offer that Silicon Valley lacks: institutional memory. They think in centuries, not quarters.
250,000 cancer papers flagged by AI as likely fraudulent. We can automate detection, but we can't automate trust — and we're not asking why the fraud market is booming.
87.4% of European SMEs don't use AI. The barrier isn't complexity or cost — it's trust. The EU is investing €1B to solve the right problem.
The EU's €1B bet on specialised AI models isn't new policy — it's validation of what already works. General models are flexible. Specialised models are actually deployable.
Deloitte used GPT-4o to write a AU$440K government report with fabricated citations and fake court quotes. The AI worked fine. The failure was human — and architectural.
Get new essays by email.