Operating principles
- 01Demos lie. Build for real users.
- 02Ship the boring parts well: the retries, the logs, the fallbacks.
- 03Measure it, or it is just an opinion.
01Aboutthe author
I build software that puts frontier models to work.
London-based engineer, self-taught by way of quantitative trading and years of building systems where the stakes were real. These days the work is applied AI: voice agents that answer live calls, infrastructure for coding agents, and defense layers that keep AI-powered support safe.
The interesting part of AI keeps moving. Models get better every few months, and each time they do, more of the hard work shifts to the software around them: the integrations, the guardrails, the parts that decide whether a clever demo survives real users.
Voice agents, coding-agent infrastructure, and AI defense layers look like different things from a distance. Up close they are the same problem: take a powerful model and build something dependable on top of it.
Current workvoice agents, coding-agent infrastructure, and AI defense layers
Biasbuild for real users, not for the demo
Backgroundquantitative trading, then self-taught systems and AI
Best fit
- 01Early systems that have to work in the real world, not just the slides.
- 02The messy seam where the model meets the product and the user.
- 03Infrastructure, applied AI, and forward-deployed work that ships to real users.