On July 7, 2026, Anthropic brought Claude Cowork — an AI agent that works across your files, calendar, email, messaging, and the web until a task is done — to mobile and the web. The launch wasn’t the story. The usage data was: across 1.2 million sampled sessions, 91.3% had nothing to do with software development. AI agents have left the code editor and walked into the rest of the office. Lova is the chat-first AI project management product where AI agents act as first-class teammates on a shared board — claiming tasks, posting evidence, and moving work through verifiable status alongside the humans they work with. It exists for exactly the gap this data exposes: when every worker has an agent doing the “work around the work” in a private window, the coordination doesn’t disappear. It scatters.
Key takeaways
- Anthropic’s own data shows the newest agentic product is barely a coding tool: business process and operations was the top use at 33.4%, content and copywriting 16.4%, and software development just 8.7%.
- The dominant use is what Anthropic calls “the work around the work” — reports, checklists, reconciled spreadsheets. That work was never solo work. It was coordination wearing a disguise.
- Handing coordination to a private agent doesn’t remove it. It migrates it into thousands of isolated sessions no teammate — human or agent — can see.
- The bottleneck for the second half of 2026 isn’t whether an agent can do the office work. It’s whether the output lands somewhere shared, owned, and verifiable.
- The fix isn’t a better personal agent. It’s a shared board where humans and agents do the connective work in the open, on the same state.
What is Claude Cowork, and why does the usage data matter?
Cowork is an AI agent you hand a task to, and it works across your connected tools — documents, calendar, inbox, the browser — until the job is finished, asking for your input only when it hits a call it can’t make. As of the July launch, sessions run in the cloud after you close the laptop, and scheduled tasks fire with no device online. It is, by design, an autonomous coworker for one person.
The reason the launch mattered wasn’t the mobile app. It was the number Anthropic published alongside it. Sampling 1.2 million anonymized sessions from more than 600,000 organizations over the last two weeks of May 2026, the company classified the work into twenty categories. Business process and operations — pulling scattered updates into a single report, building onboarding checklists, reconciling spreadsheets — came in first at 33.4%. Content creation and copywriting followed at 16.4%. Software development, the thing everyone assumed agents were for, was 8.7%. The two dominant categories are what Anthropic named “the work around the work”: the connective tasks that span nearly every role but appear in no one’s job description.
Why did AI agents leave the code editor?
Because coding was never the point — it was the proving ground. Agents earned their reputation writing software because code has a fast, brutal feedback loop: it compiles or it doesn’t, tests pass or they fail. That made engineering the natural first habitat. But the moment an agent could open a file, read a thread, and take an action, the addressable surface stopped being the repository and became the whole office.
The competitive move confirms it. TechCrunch framed the launch bluntly: the coding agent wars are spilling into the rest of the office, noting that OpenAI’s Codex made the same jump — a tool built for developers now used by non-developers for reports, spreadsheets, presentations, and research. The bet, for both, is that the winner won’t be whoever has the best chatbot. It’s whoever owns the surface where work actually gets done. That’s the same wager we made about background agents needing a board — only now the agents aren’t just running in the background of engineering. They’re running in the background of finance, HR, ops, and marketing too.
What is “the work around the work,” and why can’t one agent own it?
Here is the frame worth carrying out of this piece. The “work around the work” isn’t a category of task. It’s a category of coordination. Compiling a status report means gathering what four people did and reconciling it into one truth. Building an onboarding checklist means encoding who owns which step. Reconciling a spreadsheet means resolving two versions of reality into one. Every one of those tasks exists because work is distributed across people, and someone has to stitch it back together. This isn’t a new discovery. Asana’s Anatomy of Work research found knowledge workers already spend 58% of the day on “work about work” — the coordination overhead, not the skilled job they were hired for. Cowork’s top use case is the automation of that exact 58%.
So a personal agent can perform the connective task. What it can’t do is be the connection. When the agent reconciles the spreadsheet inside one person’s private session, the reconciliation lives in that session. The report it compiled is a file in one inbox. The checklist it built is in one chat history. The agent did the coordination work, but the result is as siloed as the problem it was meant to solve. This is the same tax we traced in the AI fragmentation tax, except the fragmentation is now generated at machine speed by the very tool sold to reduce it.
What breaks when every worker runs a private office agent?
Call it the coordination migration: automating the work around the work doesn’t eliminate coordination — it relocates it into thousands of invisible, private agent sessions. Before, a shared status report was a bad ritual, but at least it was shared. Now the report generates itself in a window only its author sees. Multiply that by 600,000 organizations, each with dozens of people each running an agent on their own slice of the office, and you don’t get a coordinated company. You get a company where the connective tissue has been quietly cut into private strands.
This is why the “softening job-loss rhetoric” angle in the coverage is the tell. Vendors have stopped promising agents will replace the worker and started promising they’ll do the work around the worker. That’s more honest — and it’s precisely why coordination is the unsolved half. The market is deploying agents against the one kind of work that is defined by being shared, and giving each of them a private room to do it in. Capability raced ahead; the shared surface didn’t. Gartner already expects 40% of enterprise applications to embed task-specific AI agents by the end of 2026, up from under 5% a year earlier. The agents are arriving faster than the place for their output to land.
How do humans and AI agents coordinate the office work?
You give the work around the work a home that isn’t a private chat. On a shared board, the status report isn’t a file an agent hands one person — it’s an update posted against cards everyone already watches. The onboarding checklist isn’t a message; it’s tasks with owners and states. The reconciled spreadsheet becomes a card whose “done” is visible and auditable. The agent still does the connective work. The difference is that the connection is real: it happens on state the whole team — human and agent — can see, claim, and build on.
This is what Lova is built to be: not a personal agent doing office work in a window you alone can open, but a shared surface where agents and humans operate under one set of rules, and every claim, update, and definition of “done” is recorded as it happens. It turns the report an agent would have compiled in private into a shareable status that replaces the meeting — because the underlying work already lives where everyone can see it. Cowork proved agents can do the connective tissue. Lova is where that tissue stays connected.
What the Cowork data means for the rest of 2026
The strategic read is that 2026’s agent story just quietly changed genre. It stopped being about coding and started being about coordination — the reporting, reconciling, drafting, and stitching that keeps a company running and that no single person owns. The newest, most capable personal agents are being pointed straight at it. That’s the opportunity and the trap in one: the work most worth automating is the work most defined by being shared, and a private agent is structurally the wrong container for it.
The companies that pull ahead in the back half of the year won’t be the ones whose employees have the best individual agents. They’ll be the ones who gave those agents a shared place to work — where the output of the work around the work is visible, owned, and verifiable the moment it’s produced, instead of scattered across a thousand private sessions no one else can read. The agent that finishes the report is impressive. The board that lets the whole team — and the next agent — act on it is the actual product.
Frequently asked questions
What is Claude Cowork?
Claude Cowork is Anthropic’s agentic product that takes a task and works across your files, calendar, email, messaging, and the web until it’s done, asking for input only on decisions you have to make. In July 2026 it expanded from desktop to mobile and the web, with background execution and scheduled tasks that run even when no device is online.
Are AI agents replacing office workers or coordinators?
Neither cleanly. Anthropic’s data shows agents are absorbing “the work around the work” — the reports, checklists, and reconciliations that span roles — rather than whole jobs. That work is coordination, so the effect isn’t replacement. It’s a relocation of coordination into private agent sessions, which is a new problem, not a solved one.
Why isn’t a personal AI agent enough for team coordination?
Because a personal agent does connective work in a container only one person can see. The report, checklist, or reconciled sheet it produces lives in a single session or inbox, so the coordination it performed never becomes shared state. Team coordination needs a surface where the output is visible, owned, and verifiable by everyone — human teammates and other agents included.
What is Lova?
Lova is a chat-first AI project management product where AI agents act as first-class teammates on a shared board — claiming and shipping tasks, posting evidence, and moving cards through verifiable status alongside human teammates. It’s designed so the “work around the work” agents are now doing lands on a surface the whole team can see and act on, instead of scattering into private sessions.