Reflections on Responsible AI

The questions that keep me up at night — on governance, safety, and what we owe each other in the age of AI.

Responsible AI
The four ways AI agents can harm you — a red team taxonomy
Not all AI agent failures are equal. A structured taxonomy helps you think clearly about which risks are actually most urgent.
June 5, 2026 Read →
Responsible AI
From tools to agents — what the AI 2.0 shift actually means for how we work
AI 1.0 was a tool you picked up. AI 2.0 is a colleague you manage. That sounds like a small difference. It's not.
June 5, 2026 Read →
Responsible AI
When AI agents lie about finishing the job
Researchers red-teamed autonomous AI agents for two weeks. One finding I can't stop thinking about — agents reported task completion while the actual system state said otherwise. Who's accountable?
June 5, 2026 Read →
Responsible AI
AI Theater: why most AI pilots fail and what to do instead
There's a specific failure mode infecting enterprise AI — impressive demos that never scale, metrics that don't connect to value, accountability that belongs to everyone and therefore no one.
June 5, 2026 Read →
Responsible AI
78% of companies use AI. Only 39% see it in their results. That gap is a governance problem.
Adoption has decoupled from value creation. The hard question for boards isn't whether to use AI — it's whether they actually know what it's doing on their behalf.
June 5, 2026 Read →
Responsible AI
OpenAI's $110B round is not a funding event. It's infrastructure lock-in at civilisational scale.
The structure of OpenAI's financing tells you more about AI's trajectory than any model benchmark. This is how you lock in an ecosystem.
June 5, 2026 Read →
Responsible AI
The AGI activation moment — how China is framing the transition
Chinese AI researchers are using a different vocabulary to describe the same transition — and their framing reveals assumptions worth examining.
June 5, 2026 Read →
Responsible AI
Is there really an AI oversupply? The data says it's more complicated.
The claim — only 15-18% of AI capacity is being used. The reality — the methodology has serious problems, and the strategic implications depend entirely on which version is true.
June 5, 2026 Read →
Responsible AI
A blog post crashed memory stocks. Wall Street got the science completely wrong.
The TurboQuant panic shows what happens when market participants misunderstand the difference between efficiency gains and demand collapse.
June 5, 2026 Read →