Claude is moving fast — new models, Claude Code releases, Agent Skills, MCP, and a whole emerging layer of tools, tricks, settings and extensions. This is the one place that maps it for business: what shipped, what it actually changes, and how your team puts it to work. Plain-language, kept current, and wired to the deeper technical leaves for whoever has to build it.
A running feed of meaningful changes across models, Claude Code and the protocols — each translated from release-note to business consequence.
Hooks, sub-agents, slash commands, MCP servers, output styles, plugins, settings — the fast-moving practitioner layer, explained without the jargon.
Designed to brief a team or a board. Every topic links to a CEO briefing for awareness and a technical leaf for the people who implement.
The themes we watch on every update. Dated, sourced entries land here as they ship — verified before they’re published, the same way every briefing is.
Tier changes, context windows, pricing and the “which model for which job” call. The decision most teams get wrong by defaulting to the biggest or the cheapest.
Technical leaf: agents/claude ↗Releases that change how work gets done: hooks, sub-agents, slash commands and skills, background tasks, IDE and web surfaces. Where the “builder is the operator” thesis becomes a real workflow.
Technical leaf: agents/claude-code ↗How capability and data get packaged and reached. Agent Skills (folders + SKILL.md) and the Model Context Protocol are the interoperable plumbing under every serious Claude deployment.
↻ Updated on a regular cadence. Each dated entry cites its primary source (Anthropic changelog, docs or announcement) and notes the business consequence — no unsourced claims.
The pieces that matter, each a maintained leaf in the knowledge tree. Start here to understand the parts before the changes make sense.
Anthropic’s flagship family — Opus / Sonnet / Haiku. Frontier reasoning, native MCP and tool-use, extended thinking. The default for production agents that need long-context reasoning and reliable function-calling.
The coding-CLI agent harness. Runs in terminal, IDE, desktop and web; orchestrates sub-agents, tools and skills. The surface where one operator commands many agents.
Claude-first agent framework — MCP-native, extended-thinking integration, sub-agent orchestration. The default when Claude is the model.
Open standard for packaging capabilities as folders (SKILL.md + scripts/refs/assets). Native in Claude Code, Claude.ai and the API; progressive disclosure keeps context lean.
Anthropic-originated, broadly adopted. The lingua franca that lets any agent consume tools and data from any MCP server.
Claude Code in context — how it compares to OpenAI Codex CLI and GitHub Copilot CLI. The honest field guide to the category.
The fast-moving layer you’re seeing emerge — the configuration and extension points that turn Claude Code from a chat box into an operated system. Deeper leaves for each are rolling out; here’s the map.
Shell commands the harness runs automatically on events — format on save, block risky edits, notify on stop. How you encode “always do X” reliably.
Specialised agents the operator delegates to — explore, review, plan — each with its own tools and context. Parallelism and separation of concerns.
Reusable, packaged workflows invoked on demand. The unit you build once and the whole team runs the same way.
Connect Claude to your tools and data — ERP, CRM, files, internal APIs — through one standard protocol, scoped and auditable.
Allow-lists, permission modes, environment config. Where safety and “don’t prompt me 40 times” are tuned for a team.
The extension and presentation layer — packaged add-ons and how responses are shaped for a given audience or workflow.
This tracker is the technical half. The CEO Briefings are the other half — the same subjects told as the decisions a leader has to make. Start with the convergence thesis (why “builder” and “operator” are the same role), then hand the relevant leaf to whoever builds it.