Project
Anthropic's agent feature in the Claude app that runs multi-step and scheduled tasks in cloud sandboxes or locally, with sessions synced across devices.
Praktika's head of talent came back from six months of leave to more than 1,000 applicants for some roles and an AI system she could not find. Business Insider's report of three returners supports planning for the gap, but shows no tested fix.
Reality
OpenAI's Help Center says Memory in ChatGPT Space can write a user's personal context onto a shared page, where every viewer can read it. OpenAI's safeguard is a review before sharing, a check that runs once while collaborators' agents keep editing.
Reality
Anthropic has merged Cowork, which has run scheduled jobs after the user closes the laptop since July, into the main Claude app. Against OpenAI's new Dots, a team's choice comes down to what starts each agent and which apps it can reach.
Reality
Claude Desktop's public JavaScript holds an undocumented setting to route agent sessions to an organization's host, a RuntimeWire review of 2,726 assets found. The code ties it to Cowork but proves neither a launch nor on-premises inference.
Reality
OpenAI opened ChatGPT Space and Pages to Pro, Business and Enterprise users at DevDay, with shared slides due within weeks. Slides export to PowerPoint and Google, so teams can work with ChatGPT's agents there and keep finished files where they already live.
Perspective Coverage
3 publishersReality
Claude agents designed 1,440 protein binders and 354 bound. The interesting part is that the prompts, provenance and every measurement went out with the number.
Reality
Instinct, an AI agent people text with, is reportedly in talks to raise $1 billion at a $10 billion valuation, according to The Information. A Wired tester credits its habit of proposing errands, but the public numbers so far count installs and investor money, with retention unreported.
Reality
Anthropic has taken the mode picker out of Claude and handed the choice to a router inside the thread. OpenAI got close to the same design in July, but it kept a toggle between Chat and Work.
Reality
Stratechery argues that slowing model improvement would shrink overhangs the labs' own speed created. The same piece says the safety motive behind Anthropic's pacing position is sincere, so both readings stand.
Reality
The company's post on trustworthy agents splits an agent into a model, a harness, tools and an environment, and says safeguards have to cover all four, at a moment when policy attention has settled on the model.
Reality
Cowork sessions now start from desktop, web, mobile and a Chrome side panel and run on Anthropic's servers with files tied to the user's account, so whether the desktop app is open decides what the agent can read on a machine.
Reality
Anthropic has renamed the three permission modes for Claude in Chrome to Manual, Auto and Skip all approvals. In Skip, by the company's own guide, nothing checks the agent's actions, and the extension remembers the choice.
Reality
Anthropic's support note says the mode picker goes away in a staged rollout, and accounts that move over cannot return to separate Chat and Cowork options. Connected apps stay live while a task runs.
Reality
Anthropic says people could not tell whether a task belonged in chat or in Cowork, so it deleted the choice. The document and deck editors arriving with the merge are beta, limited to paid plans, and built to export.
Perspective Coverage
10 publishersReality
RuntimeWire ran five assistants through the same nine tasks on September 16th with a single GPT-6 judge. All five refused the fake invoice and the lookalike payee; only Claude Cowork read the inbox the pack supplied.
Reality
The workflow gives each subagent one job and its own acceptance test, repairs what it classes as safe, and sends an uncertain semantic label back to a human with a proposed next step.
Reality
Claude Docs and in-chat slides arrive with export buttons pointing at Google Docs, Word, PowerPoint and PDF, and the sharing that makes them collaborative reaches individual Pro and Max accounts first.
Reality
KB5124008 stopped virtual machines from seeing folders shared by the Windows host, and that took local command execution away from both Windows Subsystem for Linux and Anthropic's Claude Cowork. Monday's KB5129195 restores it.
Reality
A three-month test of iOS 27 finds Siri AI's answers behind the frontier labs and its app hooks ahead of anything else on iOS. Teams planning against the assistant have two different bets to price.
Reality
RuntimeWire's static teardown of OpenAI's Codex desktop client found an onboarding card, an eligibility check that names one rival provider, and a plugin install path that waits on a server the teardown could not test.
Reality
Leadership · September 1, 2026 · Confirmed2 publishers
Product · August 25, 2026 · Confirmed4 publishers