# Knowledge hub

Notes on systems I've built, written about, or traded on — organized by domain. Credentials and shipped work live on the home page; this is where the thinking goes.

URL: https://dylanengelbrecht.dev/insights.html

## Agent architecture

- [Getting started with Cursor: automate the loop](https://dylanengelbrecht.dev/insights/getting-started-with-cursor.html): Treat Cursor as an automation surface — rules, tests, MCP, and agents run the boring parts so you keep one thread for judgment.
- [The AGENTS.md standard for AI coding agents](https://dylanengelbrecht.dev/insights/agents-md-standard.html): One open Markdown file per repo — closest AGENTS.md wins; user prompts override.
- [Organizing knowledge for AI agents](https://dylanengelbrecht.dev/insights/organizing-knowledge-for-ai-agents.html): Pair a public site with a private brain — progressive disclosure, lifespans, verification metadata.
- [Building MCP servers with search and execute](https://dylanengelbrecht.dev/insights/mcp-search-execute-codemode.html): Two MCP tools — search and execute — beat hundreds of per-operation tool definitions.
- [Reinforcement learning for tool calling in agent models](https://dylanengelbrecht.dev/insights/rl-tool-calling-composer.html): RL specializes models for the tool loop — Composer is the public case study.
- [Keep the thread: why plan–critique–build pipelines fail modern coding agents](https://dylanengelbrecht.dev/insights/single-agent-coding-loops.html): One agent, one thread, small units — verify with tests, not lossy handoffs.

## Engineering & games

- [Scalable session systems in live VR](https://dylanengelbrecht.dev/insights/scalable-session-systems-vr.html): Session architecture is where cross-platform VR becomes a reliable live service.
- [Why language models scaled first — and robotics foundation models are likely next](https://dylanengelbrecht.dev/insights/transformers-robotics-foundation-models.html): LLMs won on static data; GPU sim throughput is closing the gap for robot foundation models.

## Technical authorship

- [Why Mirror for Unity multiplayer](https://dylanengelbrecht.dev/insights/mirror-multiplayer-unity.html): Mirror is a pragmatic default when Unity teams need a complete networking architecture.
- [Progressive disclosure for learning: maps, not memorization](https://dylanengelbrecht.dev/insights/progressive-disclosure-learning-with-agents.html): Orchestrate agents across a navigable map — learn foundations deeply, index everything else, verify with the environment.

## Markets & investing

- [The Barnacle strategy](https://dylanengelbrecht.dev/insights/barnacle-strategy.html): Attach to durable capital flows; stay patient, indigestible, and leverage-free.

## Media & speaking

- [Indie showcases as proof of craft](https://dylanengelbrecht.dev/insights/rage-indie-showcases.html): Public demos are compressed systems tests — not marketing afterthoughts.