A rapid-fire tour of Claude Code's most significant recent updates: the Agent SDK for multi-agent orchestration, background subagents, LSP integration for real-time code awareness, extended thinking, and the new hooks system for workflow automation.
The Agent SDK is the headline update — it turns Claude Code from a coding assistant into an agent orchestration platform. Developers can define agent hierarchies, pass context between agents, and observe execution traces. The SDK ships with primitives for spawning, messaging, and monitoring subagents.
Background subagents run in parallel with your main session. Kick off a test suite, a PR review, or a documentation pass and get results delivered back into your workflow without blocking. This unlocks genuine multi-track development where different agents work on different concerns simultaneously.
Language Server Protocol integration means Claude reads the same diagnostics, completions, and symbol information your editor does. Errors surface before Claude runs, type mismatches are caught in context, and refactors are aware of usage across the entire codebase — not just the open file.
Hooks let you define shell commands that fire at lifecycle events: before a task starts, after a tool runs, when Claude stops. This enables automated test runs, linting passes, and integration with CI/CD pipelines without manual orchestration.
For architecturally complex tasks, extended thinking lets Claude externalize its reasoning. You can inspect the chain before it acts — catching wrong assumptions early — and tune the depth of thinking to balance speed against thoroughness.
"What shipped isn't just features — it's a new model of how you work with Claude. You're not prompting an assistant anymore. You're directing a team."