writings
notes & essays
why I disabled Claude's memory
After 2.5 years as a vocal Anthropic supporter, I disabled Claude's memory and got sharper results. Here's what changed and how I structure sessions to get more out of an LLM without letting it think it knows me.
read more →a software architect cares about the unseen things
A software architect cares about what nobody else worries about — especially when everything is working. Observability, async between components, timeouts, and maintenance mode: the unseen non-functional requirements that decide whether your system survives the day it breaks.
read more →taking responsibility is a superpower
Two conversations — one with an aerospace engineer, one with a CRO expert — pointed at the same answer to whether AI replaces experts. In the physical world, responsibility is the currency experts trade in, not just results. Software needs the same conversation.
read more →the breadcrumb trail
AI agents can't keep the whole codebase in their head, so context discipline matters more than ever. Planning has gone from the thing I neglected most to more than half my time — and the trick is documenting decisions as breadcrumbs for future sessions.
read more →your passwords are a business risk
Password management comes up in almost every engagement I have with a small business, and the situation is almost always the same. I care about this enough to write it down for anyone who wants to get it right.
read more →The AI code trust problem
AI coding assistants produce clean code faster than most engineers can type. The problem they create isn't in the code — it's in the zoom level.
read more →The rewrite trap
Most teams that come to us have already attempted a rewrite. Usually more than once. The pattern is always the same — and it fails for the same reason every time.
read more →no longer the gatekeeper
In 2023, I dropped Google for AI, the same bet I made early on Rails and Node. Two years on, developers are no longer the gatekeepers between an idea and a shipped product. Capability is the easy part. The hard skill is judgment, telling what AI genuinely changes from what it's confidently wrong about.
read more →