Anthropic Appoints Tino Cuéllar as First Chief Global Affairs Officer
Anthropic has named Mariano-Florentino (Tino) Cuéllar — a former California Supreme Court Justice — as its first Chief Global Affairs Officer. He’ll lead the company’s policy work, international engagement, and government relationships worldwide, reporting directly to President Daniela Amodei.
Cuéllar brings a heavyweight resume: a special assistant in the Obama White House, a stint on California’s highest court, and a recent run as president of the Carnegie Endowment for International Peace. He’ll take leave from Stanford Law and step down from Anthropic’s Long Term Benefit Trust to take the role. The timing isn’t subtle — the hire lands amid a Pentagon lawsuit, export-control friction, and rising tension with the Trump administration over AI policy.
Cowork’s Doubled Usage Limits End Tonight
The promotional window that doubled Cowork usage limits — run to mark the product’s expansion from desktop-only to web and mobile — closes tonight at 11:59pm PT. It applied automatically to Pro, Max, Team, and legacy Enterprise seats through August 5.
The bigger story is the surface shift underneath it. Cowork now runs on remote sessions hosted on Anthropic’s servers, saved to your account and able to keep working — and run scheduled tasks — with no device online. Anthropic’s own usage data showed most Cowork users aren’t coding, which is exactly why it’s pushing the agent onto phones and the browser where non-developers actually work.
Opus 4.1 Retires Today — and a Pricing Clock Is Ticking
Today’s the day: Claude Opus 4.1 (claude-opus-4-1-20250805) retires on the API. Anything still pinned to that model string stops answering, so point production traffic at Opus 4.8 or the newer Opus 5 before it bites.
It’s part of a wider platform tidy-up. The legacy Workbench and experimental prompt-tools APIs sunset August 17. And mark your calendar: Sonnet 5’s promotional $2/$10 per-million-token pricing ends August 31, reverting to $3/$15 on September 1. If your costs are modeled on the promo rate, rebuild the spreadsheet now.
Claude Code Makes Opus 5 the Default, Tunes Dynamic Workflows
Claude Code now sets Opus 5 as the default Opus model and drops Opus 4.7 from fast mode — /fast now applies to Opus 5 and Opus 4.8. The bundled claude-api skill defaults to Opus 5 too, with a migration path off 4.8.
On the agent side, dynamic workflows now default to a medium size guideline (aim for fewer than 15 agents), with the current default shown in the running-workflow status line. There’s also a batch of reliability fixes: killed background sessions no longer leave a permanent git worktree lock, and several long-session memory leaks got plugged — MCP stdio stderr that could balloon to 64 MB per server, and LSP documents that stayed open forever (now capped at 50 via LRU).
MCP Crosses 400M Monthly SDK Downloads
The Model Context Protocol has surpassed 400 million monthly SDK downloads — a roughly 4x jump over the year — and is now widely treated as the industry standard for wiring AI agents into applications. The open spec Anthropic seeded is no longer a Claude-only story.
The growth rides the recent MCP 2026-07-28 spec, which moved the protocol to a stateless request/response core with stronger OAuth/OIDC auth and versioned extensions for Apps and Tasks. For anyone building connectors, the takeaway is simple: MCP is where the ecosystem has consolidated, and the download curve says the momentum isn’t slowing.
Hiring a Diplomat-in-Chief Is Its Own Kind of Product Roadmap
You don’t recruit a former state supreme court justice and Carnegie Endowment president to tweak a model. The Cuéllar hire is a bet that Anthropic’s biggest constraints over the next two years are political, not technical — export controls, a Pentagon dispute, sovereignty rules, and a US administration it’s openly at odds with. Building a global-affairs function is how you keep those from becoming hard walls in front of enterprise revenue.
It rhymes with what we’ve tracked all week: in-country inference in India, governed-data deployments, a partner-certification machine. The pattern is Anthropic building the non-model scaffolding — geography, governance, and now geopolitics — that regulated buyers require before they sign. The frontier fight stopped being purely about benchmarks a while ago. Increasingly it’s about who can operate inside the rules, and who can help write them. Today Anthropic hired someone to do both.