🧭 Claude Cowork Expands to All Pro Subscribers
After launching in a Max-only research preview on January 12, Anthropic has opened Claude Cowork to all Pro plan subscribers today. Cowork is a desktop agent for macOS that breaks from the chat metaphor: rather than waiting for turn-by-turn replies, it works autonomously inside a user-designated folder, completing tasks in parallel and pausing only when a decision genuinely requires human input. Anthropic describes the experience as "leaving messages for a coworker rather than prompting an assistant."
What Pro subscribers can do today
- File I/O natively — read, write, and edit documents, spreadsheets, and PDFs in the designated workspace folder
- Parallel task execution — Cowork spawns sub-tasks that run concurrently, aggregating results before surfacing output
- Office document export — produce XLSX, PPTX, DOCX, and PDF artefacts from natural-language instructions
- Browser integration — pairing with Claude in Chrome unlocks web-based task completion as part of a workflow
Getting started: Designate a dedicated workspace folder (not your entire home directory) and start with a clearly scoped, single-output task — for example, "summarise all PDFs in /reports/ into one executive summary." Cowork works best when it has clear success criteria and a bounded input set.
Cowork
productivity
macOS
retrospective
🧭 DSIT Selects Anthropic to Build the GOV.UK AI Assistant
The UK's Department for Science, Innovation and Technology has formally selected Anthropic to build a Claude-powered AI assistant for GOV.UK, Britain's central government website. The assistant goes beyond FAQ retrieval: it actively guides citizens through government processes, with the initial focus on employment — finding work, accessing training, and understanding available benefits. Accuracy improved from 76% to 90% across two prior pilot phases covering more than 26,000 questions, clearing the threshold needed for a national rollout.
The deployment builds on the MOU Anthropic signed with the UK government in early 2025 and marks the first national-scale government deployment of Claude for a public-facing citizen service. The pilot will operate initially in English, with Welsh and other language support planned for later phases. GOV.UK serves approximately 60 million unique users per month.
Accuracy benchmark context: The 90% accuracy target was measured against a curated dataset of citizen queries with verified correct answers — comparable to, but distinct from, standard LLM benchmarks. Government deployments typically demand higher accuracy floors than consumer applications because errors carry real procedural consequences for citizens.
government
UK
public sector
retrospective
🧭 MCP Tool Search & Programmatic Tool Calling Reach General Availability
Anthropic has moved two major API capabilities out of beta today: the Tool Search Tool and Programmatic Tool Calling (PTC). Neither requires a beta header any longer, and both are available across all model tiers.
Tool Search Tool
Rather than loading hundreds of MCP connector schemas into the context window upfront, Claude can now discover tools on-demand from a library at query time — preserving up to 85% fewer tokens versus loading the full tool list. This is particularly valuable for enterprise deployments with large connector libraries, where token overhead was previously a production concern.
Programmatic Tool Calling
PTC lets Claude write and execute code that calls multiple tools in a single round-trip inside a sandboxed container. On multi-step agentic workflows — retrieve data, transform it, call another API — the latency improvement is significant. Code execution is free when bundled with web search or web fetch calls under the new pricing structure.
API
tool use
MCP
developer tools
retrospective