Writing
Thoughts on engineering, infrastructure, and building systems.
The unit of engineering output is no longer the person. It's the person plus the harness they've built to direct a fleet of agents, and the leverage is real, right up to the ceiling where judgment refuses to scale.
Agents optimize what they can verify: does it render, do the values map, do the tests pass. They cannot optimize what they can't measure, whether the result is legible, calm, honest, good. That gap is where the human reviewer's job moved.
AI-led migrations are a genuine power tool. They sweep wide, mechanical changes across a codebase at a speed a solo developer can't touch. They also migrate the happy path and leave the rest to rot quietly, unless you design the migration to prove itself rather than just finish.
AI doesn't lower the knowledge bar for the person directing it. It raises it. The case for continual learning in an age that keeps insisting you'll never need to learn anything again.
The leverage in agentic development isn't any single library. It's the stack of primitives you build underneath, where each layer makes the next one obvious and every project subsidizes the one after it.
Agentic tools nail the easy 80 percent. The last twenty, comprehension and reliability and judgment, still belongs to you. The next frontier moves down the stack, into the silicon.
First post on the new site. A fresh start with Astro, Tailwind, and a focus on content over flash.
Why I built a lightweight Go library for durable job queues with checkpointed workflows, crash recovery, and an embedded monitoring dashboard.
How I built a type-safe Go SDK for Langfuse's LLM observability platform, covering the batch processor, hierarchical tracing, and API design decisions.
No posts found