Agent platforms and standards this week: Duet, Skills→AAIF, ACE, Decisions API, and formal proofs
Anthropic shipped Claude Duet research preview and donated Agent Skills to the AAIF; Google launched Agent Compute Engine; OpenAI’s Decisions API went GA; Anthropic also posted assembly-level no-crash formal-verification results. Product preview, open standard, infrastructure and proofs are related tracks—not interchangeable.
Anthropic opened Claude Duet as a research preview: standalone iOS/Android apps and claude.ai on Claude Opus 5.5, with morning briefings and background tasks, and confirmation before send, pay, sign or delete. The US Pro/Max preview counts against usage with no extra fee. That is an end-user proactive-agent product preview—not the open standard itself.
Separately, Anthropic donated the Agent Skills open standard to the Agentic AI Foundation (AAIF) under the Linux Foundation, citing 70+ agent platforms, 1 million skills published and nearly 280 million installed since launch, and transferring the name, logo, GitHub org and agentskills.io. That is a governance/interoperability handoff—read it apart from the Duet product preview.
Google Cloud at AGNTCon+MCPCon introduced Agent Compute Engine: fast start/stop hardware-isolated microVMs per agent, already backing Gemini work tasks, with customer preview expected in coming months and a design that avoids long-lived credentials inside the sandbox. OpenAI made the Decisions API generally available for low-latency classify-and-route answers, claiming ~10× the Responses API. One is cloud execution sandboxing; the other is an application decision API—neither is “just another chat agent.”
Anthropic, with Theorem, said Claude wrote machine-checked no-crash proofs for 123 real open-source libraries at RISC-V assembly level, verified by an external proof checker, and stressed that formal verification is not the same as “no bugs.” That is a software-correctness proof track with no direct causal link to Duet, Skills or ACE—kept here as a same-week contrast.
Sources
- Claude Duet — Anthropic
- Donating Agent Skills to the Agentic AI Foundation — Anthropic
- Introducing Agent Compute Engine and expanding agent governance — Google Cloud
- Decisions | OpenAI API — OpenAI
- LLMs are making formal software verification feasible at a much larger scale — Anthropic