Writing
Blog
Thoughts on software engineering, AI, fintech, and building things that matter.
The Through Attack: The AI Security Threat Nobody's Naming
Two attack directions dominate AI agent security. There's a third — the Through attack. It doesn't touch your files. It controls what your agent believes. And it's already in the wild.
The 43-Point Gap
Experienced developers using AI were 19% slower — but believed they were 20% faster. That 43-point perception gap explains more about the state of AI-assisted development than any benchmark.
Velocity Is a Feedback Loop
Most teams think velocity means shipping more. In practice, velocity is how quickly you learn what to stop building.
Is This the Product, or Is This the Machine?
Every piece of agent infrastructure feels like it serves the product — until you step back and realise you've spent two weeks serving the pipeline.
Clever Is Not a Feature
The best code I've written wasn't elegant. It shipped, it worked, and users never had to think about it. That took me years to understand.
The Mirror Problem: Self-Referential Loops in Modern AI
AI is writing code that trains AI. AI is judging AI outputs. AI is scraping AI-written content to learn from. Here's what the research says happens next.