Anthropic is done building an AI that hands you a rough draft and wishes you luck.
On Tuesday, the company announced a major expansion of Claude Cowork, its AI agent platform, with a new plugin system that connects Claude directly to the enterprise tools people actually use: Salesforce, Slack, Excel, PowerPoint, Google Workspace and more. Instead of copying Claude's output into another app, Claude now reaches into those apps itself.
"Cowork makes it possible for Claude to deliver polished, near final work," Scott White, Anthropic's head of product for Claude Enterprise, told VentureBeat. "It goes beyond drafts and suggestions — actual completed projects and deliverables."
The company was unusually candid about why this matters now. "2025 was meant to be the year agents transformed the enterprise, but the hype turned out to be mostly premature," Kate Jensen, Anthropic's head of Americas, said. "It wasn't a failure of effort, it was a failure of approach."
The new approach is plugins. Think of them as bridges between Claude and your work tools. Rather than you being the middleman, Claude crosses the bridge itself, pulling sales data, updating project boards, building slide decks. The initial batch covers the usual enterprise suspects, with more coming.
There's already evidence this isn't just a developer tool. At Epic, the healthcare tech company behind MyChart, over half of Claude Code usage now comes from non-developer roles, according to Seth Hain at the company. That's a telling sign of where Anthropic's tools are headed: further from the engineering floor, closer to everyone else.
These agents are also working longer without checking in. Anthropic's own research on agent autonomy found that the longest Claude Code work sessions nearly doubled between October 2025 and January 2026, from under 25 minutes to over 45 minutes. Not long enough to manage your whole day, but the trajectory is steep.
Longer sessions and deeper tool access naturally raise questions about accountability. "Traditionally, we are so used to building deterministic platforms. You write code requirements and build. And now, with AI being probabilistic, the accountability doesn't go away," Sridhar Masam, CTO of the New York Stock Exchange, said.
Anthropic says its guardrails are keeping pace. Its research found that 80% of agent tool calls include at least one safeguard, and only 0.8% of actions appear irreversible. One notable finding: Claude asks for clarification on complex tasks more often than humans choose to interrupt it. The model is more cautious than its users.
Still, the technology may be outrunning the organizations trying to adopt it. "The tools are in many senses ahead of the change management," Steve Hasker of Thomson Reuters noted. A general counsel's office or an audit firm needs to rewire how it works before it can truly take advantage.
Valley View
Anthropic's real play here isn't the plugins. It's becoming the intelligence layer between you and every tool you use at work, a platform move that, if it sticks, makes Claude as hard to leave as your email client. The question is whether organizations can rewire their processes fast enough to keep up with what the technology already offers. JPMorgan recently assessed that full agentic replacement of enterprise software is a post-2028 story at the earliest. Anthropic is betting it can get you hooked well before then.
