Stay updated on the latest in AI agents, software systems, and how teams are bridging deep domain expertise with the engineering loop. New essays, walkthroughs, and case studies — every other week.
Most "AI code reviewers" collapse the whole job into one big prompt. We argue for a small, opinionated graph of specialists — and show what that bought us inside MindPortalix.
Six months in production with a fintech engineering org. We share what changed in the review loop, what kinds of bugs got caught earliest, and what the team kept doing manually.
A walkthrough of how domain experts onboard knowledge into MindPortalix using plain-language workflow files, references, and MCP tools.
A teardown of the extension's transport layer, why we picked SSE over WebSockets, and the trick that shaved 1.6s off cold starts.
An honest review of the Model Context Protocol after building 12 production agents on top of it. What it solved, what we still hand-roll, and what we'd build next.
A short retrospective on the kinds of problems we keep being pulled toward, and the working theory behind why a "portal" between roles works better than yet another doc tool.
Not every commit deserves a check. We outline the five events where domain feedback consistently changes outcomes — and three you should leave alone.
A pragmatic eval setup for teams shipping agentic systems — golden cases, regression suites, and the cheap signals we trust before more expensive ones.
Join the SERVMIND community for early access to product changes, design notes, and a Slack channel full of people building agents that respect the work.