Recent writing

Better at what, and for whom? Keeping AI useful and under human control has to be part of what we mean by improvement.

Sam, Dario and Sacks are making consequential decisions about pace, oversight and whose risks count.

Why working software, not documentation or ceremony, is the artifact that actually reduces uncertainty — and why that lesson still holds in the AI era.

Reflections from an Anthropic research interview on why language models need honest framing, and why human agency has to stay at the centre.

Hearing Messiah live again sent me down a rabbit hole of counterpoint, class, and why this very Protestant oratorio still works in 2025.

After fourteen years as a loyal Spotify customer, I finally cancelled my subscription. Here's why the platform's algorithmic direction no longer matches my simple, self-curated listening style.

Exploring how penetration testing has evolved alongside modern software development practices, and why AI-assisted red teaming might be the future of security validation.

We're not 6 months from AGI; we're 6 months from confronting the security implications of AI systems with real-world agency.

The tech industry has shifted from custom builds to piecing together existing components - it's all just plugging Legos now.

I thought I was immune to horror games. Then Resident Evil made me take the headset off and have a sit down.