Third Protocol
My views on Tech, AI governance and employee wellbeing
The risk of handing over control to AI
The risk that concerns me most is how easily we are handing control to the companies that own AI.
EU AI Act
View allThe EU AI Act deadline moved
The main deadline for high-risk AI systems is now 2 December 2027, but some of the Act already applies and the first date to plan for is 2 December 2026.
AI governance
View allWho pays when an AI agent gets it wrong?
Your AI agent buys the wrong thing and the shop won't refund it. AI can't be held liable in law, so the answer is always a person or a business.
The UK's rules on automated decisions
There's no AI Act in the UK, but since February 2026 UK GDPR has set four things you must be able to do when a system decides about a person on its own.
Who reviews the code reviewer?
If automated review were reliable enough to trust alone, the code it reviews would be too. It ends when someone takes responsibility.
Guardrails in practice
View allSupervision is not a control
A person watching an AI agent doesn't stop it. Limits set in advance do.
Verification debt
When a machine writes the code, you have to build your understanding of it during the review. Werner Vogels called it verification debt.
Notes on reviewing AI-written code
An agent doesn't learn from a code review, AI code looks finished before it is, and approval prompts stop being read.
Views
View allThe risk of handing over control to AI
The risk that concerns me most is how easily we are handing control to the companies that own AI.
An A–Z of software engineer wellbeing: D to F
Deadlines, expectations and fatigue. The next three letters of an A–Z on what this job does to the people doing it.
An A–Z of software engineer wellbeing: A to C
AI anxiety, burnout and context switching. The first three letters of an A–Z on what this job does to the people doing it.