Who is building the agent operating system?
Twelve dimensions across seven providers. Re-graded as announcements land, with the market's reaction shown alongside.
Anthropic · On-Device AI
Privacy and latency · updated 2026-06-01
No device or local-inference presence at all.
How much capability runs locally, for privacy, latency, offline use, and cost? Hardware, quantized models, and the OS-level plumbing that keeps sensitive context off the server.
No outside coverage tagged to this exact cell yet. Here is recent Anthropic analysis.
- NewsAnthropic says Claude can now use your computer to finish tasks for youCNBC · 2026-03-24
Anthropic expands Claude's computer-use as part of a push toward agents that operate screens and apps for you.
- AnalysisAgents Over BubblesStratechery · 2026-03-16
The AI value chain shifts to integrated model-plus-harness systems whose orchestration creates defensible moats.
- AnalysisA Guide to Which AI to Use in the Agentic EraOne Useful Thing (Ethan Mollick) · 2026-02-18
A models/apps/harnesses framework: the harness that lets a model take real actions now matters as much as IQ.
- MarketLeaders, gainers and unexpected winners in the Enterprise AI arms raceAndreessen Horowitz (a16z) · 2026-01-30
An enterprise survey: OpenAI still leads adoption but Anthropic and Google gain fast, a multi-vendor oligopoly.
- AnalysisAnthropic's Claude Code is having its 'ChatGPT' momentUncoverAlpha · 2026-01-26
Claude Code's terminal-native agent design drove rapid enterprise adoption and measurable productivity gains.