Why six AI labs built the same product for knowledge workers in four months
Six vendors built the same agent in four months.
Anthropic shipped Claude Cowork on January 12, gave it Claude Code’s agentic harness, and watched investors knock $285 billion off the SaaS index three weeks later when the open-source plugin pack landed on GitHub.
Perplexity launched Computer on February 25, an orchestrator that routes work across nineteen models with Claude Opus as the reasoner.
Microsoft announced Copilot Cowork on March 9, built on Claude through its deepening partnership with Anthropic, with Frontier preview rolling out in late March.
OpenAI rebuilt the Codex desktop app on April 16 with computer use, ninety plugins, persistent memory, and scheduled automations, and Greg Brockman called it a general agent harness that happens to write software.
Google launched Gemini Enterprise Agent Platform and Workspace Intelligence at Cloud Next on April 22, with an inbox of long-running agents and a daily briefing that drives Docs, Drive, Gmail, and Chat.
Amazon launched the Quick desktop app on April 28, complete with a personal knowledge graph, background monitoring, and connectors for Google Workspace and Microsoft 365.
Six vendors. One pitch: The agent works alongside the knowledge worker, reads local files, drives the browser, retains context across days, and delivers finished outputs rather than suggestions.
The convergence has a clear cause
Claude Code proved that agentic harnesses on top of frontier models could ship real work. Every lab that watched developers fall in love with Claude Code asked the same question. Why should this stay a developer tool? Anthropic’s own answer was Cowork, and Kate Jensen told CNBC the bet is that every knowledge worker will feel about Cowork the way engineers now feel about Claude Code. Microsoft, OpenAI, Perplexity, Google, and Amazon are running the same play.
Knowledge workers, though, are not developers. The Claude Code adoption curve assumed an audience that already lived inside a terminal, already understood what a file system was, and already knew how to read an error message. The Cowork pitch asks marketing managers, finance analysts, HR leads, and operations teams to do something engineers spent two years learning. Delegate a multi-step task. Supervise an agent doing the work. Catch the moment it goes off the rails. Approve actions before they are sent. Trust an output that was not produced one keystroke at a time.
The gap is behavioral, not technical.
The early evidence is mixed. Microsoft’s April earnings call disclosed twenty million paid Copilot subscribers, up from fifteen million in January. A 33 percent jump in a single quarter. Real momentum, but still under 5% of the 450 million commercial Microsoft 365 base. PwC committed in April to rolling out Cowork and Claude Code to hundreds of thousands of professionals worldwide, the largest enterprise cowork deployment announced so far. The supply side has buyers. Whether those buyers translate seat licenses into daily habits is the part nobody can see yet.
The contrarian read is that the AI-assisted work wave will look more like a long enterprise rollout than a viral consumer moment. Adoption will track approval workflows, audit trails, and change management timelines. The vendors who win will be the ones who align with the existing rhythm of how knowledge workers already work, not the ones with the slickest demo. Microsoft has the rhythm. Google has the Workspace context graph. Amazon has the data graph. Anthropic has the harness. Perplexity has the orchestration. OpenAI has the brand and three million weekly developers it can drag forward.
The supply side has buyers. Whether those buyers translate seat licenses into daily habits is the part nobody can see yet.
The version of this story I am most likely to be wrong about is the one where the PwC rollout produces a viral internal demo, and the procurement curve compresses from quarters to weeks.
The agents are ready. The question is whether the people they were built for are.
