Now
A snapshot of what has my attention at the moment — what I am building, learning and reading. Inspired by the /now movement; kept honest rather than complete.
Last updated:
Building
- A customized AI-assisted coding workflow — agents, prompts and guardrails I lean on across every project — with a close eye on what genuinely improves productivity versus what just looks clever.
- Helping the developers I lead adopt new tools and build their own AI workflows, rather than inherit mine wholesale.
- This site — a small playground for typography, motion and content architecture in Astro.
Learning
- Context engineering for agents: how much to retrieve, when to summarize, and how to keep a long-running agent from losing the plot.
- Astro's content layer and view transitions — enough to rebuild this whole site on them and genuinely enjoy the process.
- Prompt caching and streaming patterns — the practical mechanics of making LLM calls fast and cheap enough to ship.
Reading
- Papers and write-ups on evaluation for LLM systems — the unglamorous, essential half of actually shipping AI.
- Revisiting Domain-Driven Design with fresh eyes — a few more years of scar tissue make the ideas land differently.
- Other engineers writing about their own tools and workflows — half my good ideas start life as someone else's.
Exploring
- Local and open-weight models — how far you can get without a frontier API, and where the quality cliff actually sits.
- Typed, testable prompts — treating an LLM call like any other function with a contract, not a magic incantation.
Recently discovered
- How much of an agent's reliability comes from good tools and tight feedback loops, not a cleverer model.
- Modern CSS — color-mix(), container queries, :has() — has quietly removed most of my reasons to reach for a preprocessor.
Thinking about
- Where a human belongs in an automated pipeline — which exceptions are worth a person, and which are just noise.
- How to write software that an AI assistant and a human can both read comfortably.