Anthropic's furious sprint
Anthropic has pushed 72 releases in 52 days — a cadence that helped its ‘hectocorn’ valuation jump +7.72% yesterday. The rollout includes massive capabilities like a 1M‑token context window, voice mode, agent teams, scheduled tasks, persistent memory and mouse/keyboard computer control — positioning Claude as an “AI engineer” teammate. (x.com) (x.com)
Anthropic published the Opus 4.6 upgrade on Feb. 5, 2026 and shipped an experimental “Agent Teams” workflow that coordinates up to multiple independent Claude Code sessions via a shared task list and direct inter‑agent messaging. (help.apiyi.com) On March 13, 2026 the company removed its long‑context surcharge and made the full million‑token capacity for Opus 4.6 and Sonnet 4.6 generally available at standard per‑token rates, a pricing change Anthropic framed as lowering a barrier for long‑document and multi‑session use. (winbuzzer.com) A voice interaction beta arrived in mid‑March 2026, enabling two‑way spoken conversations on web and mobile with selectable voices, push‑to‑talk and hands‑free modes, and automatic transcript saving to chat history. (agent-wars.com) Claude Cowork launched as a macOS desktop research preview on Jan. 12, 2026 and Anthropic added a “Dispatch” mobile‑to‑desktop preview on March 17, 2026 that lets users trigger and monitor desktop agent tasks from a phone via a persistent conversation thread. (venturebeat.com) Claude Code gained a cron‑style /loop command in early March 2026 that runs recurring background jobs while a session is active, with runs that auto‑expire (Anthropic’s docs and hands‑on reports note a three‑day auto‑delete behavior for some background tasks). (code.claude.com) Anthropic’s Memory Tool, introduced in beta in late September 2025, exposes a file‑backed persistent memory directory for agents to create, read, update and delete long‑lived memory files, and the feature was expanded to paid plans with import/export controls later that year. (platform.claude.com) Mid‑March reports and Anthropic documentation show the macOS desktop agent features (including simulated mouse/keyboard control and sandboxing) arrived as research previews for paid tiers while the company concurrently published advanced patterns for subagents and MCP integrations aimed at scaling agent deployments. (appleinsider.com)
