navigator.modelContext still returns undefined, but behind that word hides a new way for the web to expose tools straight to an agent — no server. Turn on the flag, register a tool, and test it.
Anthropic shipped the /goal slash command. A session-scoped wrapper over a prompt-based Stop hook: you write a condition, a small evaluator reads the transcript after each turn, and the session ends only when it says yes. When to reach for /goal, when for /loop, when for a Stop hook — and why you write the condition for Haiku, not for the main model.
My bet: by spring 2027, isolated context per sub-task gets a folk name and jumps from engineering blogs into pitch decks. Why the Lütke–Karpathy–Anthropic pattern keeps repeating.
Agents deliver sloppy code because they get sloppy briefs. A practical recipe for turning a Jira ticket into binary acceptance criteria, three-tier boundaries and self-verification.
A talk on AI agents in software development. CLAUDE.md, hooks, skills, Skillsmith, git worktrees and real-world case studies — everything I showed at Yolk Studio.
The Claude Agent SDK gives you the same primitives Claude Code uses — tools, hooks, subagents, permissions. A practical walkthrough of shipping one to production.
AI agents waste tokens parsing your navbars and cookie banners. I built an open-source library called next-markdown-mirror that fixes this — and it costs $0.