Cua Driver Bench Tests Whether Desktop Agents Can Work While You Keep Typing
Cua's benchmark separates task completion from undisturbed background work. Its reported macOS results show why desktop-agent builders need to measure both.
We use cookies to measure how the site is used and how our ads perform (Google Analytics and Ads, Microsoft Clarity, and the Meta, OpenAI and X ad pixels). “Essentials only” turns off advertising cookies everywhere, and analytics too in the EEA, UK and Switzerland. Privacy Policy
AUTOMATER NEWSROOM
New tools, useful workflows, and consequential AI developments.
Subscribe via RSS ↗OpenAI's Codex agents autonomously investigate and fix ChatGPT speed regressions at scale: 1,700+ investigations and 600+ root causes in three months.
Job listings spotted by Bloomberg show Apple building an autonomous agent that browses, codes, controls apps and coordinates sub-agents — the last major platform holder to enter the computer-use race.
Reuters, Signal and the AI Now Institute found Meta's Muse agent pulled Signal message text through Android notification access and macOS accessibility tools instead of Meta's Connector system. Staff also reported data that wasn't theirs.
Cua's benchmark separates task completion from undisturbed background work. Its reported macOS results show why desktop-agent builders need to measure both.
A verified GitHub issue reports Claude Code sent the signed-in account email in User-Agent headers. A user documents model-composed paths the default v2.1.317 privacy mod may miss - here is a bounded fleet check.
ToyBlocks combines automatic demonstration generation, policy training and controlled evaluation, with 85,336 long-horizon demonstrations and initial physical-robot tests.
OpenRouter rebuilt its router and shipped openrouter/auto-code for coding agents: eleven request factors, live price, health and cache signals, a 3x faster routing step than the previous router, and 74% lower cost per task than Claude Opus 5.5 on OpenRouter's own five benchmarks.
@GitHub is generally available in Slack and Teams. Teams Canvas adds shared implementation plans, code changes and previews, while the announcement leaves pricing and permission details unspecified.
Codex CLI 0.167.0 ships a no-daemon agent command center, backend switching that keeps drafts, composer image previews and a Vertex AI setup wizard.
Qwen Code v0.26.8 adds Qwen subscription support across the CLI, VS Code extension and Desktop (#16378), with concurrent managed-agent workspace rewinds and GitHub links in channel events. What the release establishes, and what it leaves unverified, about accounts and billing.
AWS moved Web Answer on Amazon Bedrock AgentCore to general availability on Nov. 5, 2026. One MCP tool call returns a synthesized, cited answer for agents.
Cursor's Nov 5 changelog ships Build Live Apps: a cloud agent scaffolds the app, provisions Postgres, auth and hosting, four always-on security checks gate the publish, and the result is a link your team or a script can open. What shipped, what the gate blocks, and the admin switches to set first.
Google launched Gemini Omni 1.2 Flash with 15-second clips, conversational editing, and a 30% price cut to $0.07 per second for 720p video on Google Cloud.
Anthropic is rolling Personal Finance out in beta to U.S. Pro, Max, and Ultra subscribers. The tool connects bank and card accounts through Plaid, answers from balances and up to two years of transactions, and cannot move or invest the money.
Xcode 27.3 beta adds GitHub Copilot as a backend for Apple's built-in Coding Intelligence panel, in public preview alongside ChatGPT, Claude and Gemini.
Claude can now sign MCP connector logins itself, skipping per-app OAuth for Team and Enterprise orgs without an identity provider. Beta covers 14 connectors.
Menlo's 2026 State of Enterprise AI report puts deployment spend at $176.8B, coding at $34B, Claude Code and Codex at $24B, and average spend at 2.5x budget.
Google signed three credit support agreements backstopping Anthropic's 401 MW, 20-year leases at TeraWulf's Hawesville, Kentucky campus. How the backstop works, what ends it, and what is still unfinished.
HumanLayer is joining Factory. The hosted platform stays up for a transition period, and users can export, move to Factory, or self-host the open-source code. Here is what is confirmed, what is only reported, and what to do first.
OpenAI reports an autonomous pipeline reproduced 1,724 Linux kernel vulnerabilities and produced patches for 948 local privilege escalation flaws at $1,213,430 in API credits. What changes for operators: patch velocity, sandbox isolation, revalidation and adversarial patch review now carry the threat model.
Meta has acquired Conductor, the Mac app for running Claude Code, Codex and Cursor agents in parallel. Conductor Cloud goes read-only Nov. 13 and shuts down Nov. 20. The local app gets new models and critical fixes until early 2027.
Postman desktop auto-installed Claude Code and Codex plugins for existing users, and the installer code attempted to approve its own Codex hook. A GitHub-thread reply said Postman is reviewing that flow.
Moonshot AI released Kimi K3.1 Pro open weights: a 2.8T-parameter, 104B-active, 1M-context multimodal agent model under a custom license. vLLM shipped day-0 serving on NVIDIA, AMD, TPU and Alibaba PPU hardware, plus a zero-mismatch RL path.
Our AI-assisted editorial method · Explore our in-depth Intel guides
Get started
Lite is free forever, installs in about a minute, and reads the session history you already have.
Also for macOS (Apple silicon) and Linux.
Signed installer. No account. Session content stays on your disk.