Writing from the team
Notes on repo context, coding agents, and how we think about building Driftless.
┌──────────────────────────────────────────────┐ │ ▌ repo.context rev 0001 │ ├──────────────────────────────────────────────┤ │ │ │ ▸ context lives with the codebase │ │ └─ updated by every merged PR │ │ │ │ ▸ delivered before changes are made │ │ └─ not after review catches them │ │ │ │ ! gaps grow faster than teams can track │ │ └─ risk: high │ │ │ │ → PR observation loop: continuous │ │ └─ decision │ │ │ ├──────────────────────────────────────────────┤ │ applies to all files · owner @team │ └──────────────────────────────────────────────┘
AGENTS.md best practices: how to brief AI coding agents (2026)
Eight practices for an AGENTS.md that actually improves AI agent output — what to put in, what to leave out, and where deeper per-file context belongs.
Best ways to give AI coding agents context about your codebase (2026)
Six proven ways to brief Claude, Cursor, Copilot, and Codex on your codebase — AGENTS.md, editor rules, MCP, RAG, comments, and a code-anchored context layer. Honest trade-offs.
MCP vs. a context layer: connecting agents to tools isn’t the same as giving them context
MCP connects AI agents to tools and data. A context layer decides which knowledge a change requires. They solve different problems — and a serious setup needs both.
Why AGENTS.md isn’t enough for AI coding agents
AGENTS.md briefs an agent on your repo — once, globally, for the whole tree. Real codebases need per-file context that changes as the code does. Here’s the gap and the fix.
Context drift is the hidden tax on AI-assisted teams
Every stale note your team trusts is a tax paid in broken changes and re-derived knowledge. AI-assisted coding raises the rate. Here’s how to stop paying it.
Topics are now notes in Driftless
Driftless topics used to be rigid forms. Now they are markdown notes with tags, frontmatter, and full flexibility. Here is why that matters.
Driftless is here: context delivery for AI-native teams
How we went from "read the code" to "here is the context you need." What Driftless does, what you can do today, and why context delivery matters for AI-native teams.