Anthropic's official "Code with Claude" workshop materials — nine hands-on courses on skills, MCP, evals and Claude Managed Agents.
- official
- workshops
- anthropic
- managed-agents
Anthropic Engineering Blog
Anthropic's engineering blog — how Claude Code, agents, and harnesses are actually built.
Anthropic's official engineering blog, publishing first-party deep dives on building with and for Claude: agent harness design for long-running development, Claude Code's auto mode, scaling managed agents, and similar practitioner-grade posts. The primary source for Anthropic engineering thinking beyond product docs.
What it does
- First-party engineering deep dives from the team building Claude products — the tagline is literal: how Claude Code, agents, and harnesses are actually built
- Recent run covers the infrastructure questions: how they contain Claude across claude.ai, Claude Code and Cowork (capping the blast radius of more capable agents); scaling Managed Agents by decoupling the brain from the hands; and the design of Claude Code auto mode as a safer way to skip permissions
- Harness design thread includes harness design for long-running application development and effective harnesses for long-running agents
- An unusually strong eval streak: eval awareness in Opus 4.6 BrowseComp performance, quantifying infrastructure noise in agentic coding evals, designing AI-resistant technical evaluations, and demystifying evals for AI agents — plus a C compiler built by a team of parallel Claudes
When to use it
The primary source when you want agent-infrastructure design thinking straight from a production team — read it before designing your own harness, eval suite, or containment model.
Getting started
No install — a blog. Posts carry dates (the archive spans 2024-09 to 2026-04); read the newest first.
Things to know
Posts reference the model generation they shipped with — mine them for patterns and tradeoffs, not for version-specific behavior that newer models may have changed.
Charted
Related on the map
Anthropic's announcement of Fable 5.1 and Mythos 5.1 — one base model, two guardrail postures.
- official
- models
- announcement
- fable
Anthropic's canonical essay on agent design — workflows vs agents, and five composable patterns.
- official
- agents
- engineering
- patterns