What an agent and a harness are
- · I think "agent" may finally have a widely enough agreed upon definition · Simon Willison · The short definition the later pieces build on: tools in a loop toward a goal.
- · The Anatomy of an Agent Harness · Viv Trivedy · Where "Agent = Model + Harness" comes from. About 20 min.
- · Building Effective AI Agents · Anthropic · Anthropic's base text on agents versus workflows.
Harness engineering
- · Skill Issue: harness engineering for coding agents · HumanLayer · The practical levers: AGENTS.md, skills, sub-agents, hooks and back-pressure. About 20 min.
- · Agent harness engineering · Addy Osmani · A survey that draws the other harness pieces together. About 15 min.
- · Harness engineering for coding agent users · Birgitta Böckeler · Guides and sensors: a frame for the controls around a coding agent.
- · My AI Adoption Journey · Mitchell Hashimoto · A practitioner's account.
Context and memory
- · LLM Wiki · Andrej Karpathy · A pattern for a plain-markdown wiki that an agent keeps and reads. About 10 min.
- · Effective context engineering for AI agents · Anthropic · On choosing what goes into an agent's context.
- · Context Rot · Chroma · The research behind "longer context, worse results".
- · 12-factor agents · Dex Horthy, HumanLayer · Principles for building LLM-powered software.
- · Memory overview · LangChain · The short-term and long-term memory terms, defined.
Work order: what starts, how many at once
- · Bets, Not Backlogs · Ryan Singer, Basecamp · Shape Up on betting on shaped work instead of keeping a backlog.
- · The Betting Table · Ryan Singer, Basecamp · How a team decides what to bet on.
- · Write the Pitch · Ryan Singer, Basecamp · How to write the pitch that gets bet on.
- · Embracing the parallel coding agent lifestyle · Simon Willison · Running agents in parallel, where review is the limit.
- · The bottleneck AI coding assistants moved rather than removed · Stavros · Review, not writing, is the limit.
- · WIP-1, Little's law and agent throughput · Agent Patterns · One task open per agent until it is verified.
Long-running agents and review
- · Effective harnesses for long-running agents · Anthropic · Keeping an agent on track across long-running work.
- · Scaling Managed Agents · Anthropic · Running hosted agents at scale, the headless route.
- · Code Review for Claude Code · Anthropic · A model reviewing every pull request, with its numbers.
- · From Industry Claims to Empirical Reality: Code Review Agents in Pull Requests · arXiv 2604.03196 · 3,109 agent PRs: why a human still approves the merge.
Tooling for agents
- · ViteConf 2026 livestream · Evan You · "Developer Tooling in the Agentic Era", from 0:03:30, about 25 min: the Vite Plus direction.