Claude Can Write the Offer Letter. It Can't Tell You Who Owns the Agent That Did.

On February 24, 2026, Anthropic announced that Claude Cowork would expand to 13 enterprise plugins spanning HR, design, engineering, operations, financial analysis, investment banking, equity research, private equity, and wealth management. The announcement was framed correctly: this is enterprise AI moving from experimental to infrastructural.

For HR specifically, the plugin enables job description generation, onboarding materials, and offer letter drafting — connected to tools like Google Drive, Gmail, and DocuSign. It's a real workflow improvement. Most HR teams that have used any AI for these tasks will recognize what the plugin does and why it reduces administrative load.

The software market recognized the moment. Salesforce rose 4%. Thomson Reuters, whose FactSet product was named as one of the integration partners, climbed 11%. The message investors received was: enterprise AI automation is accelerating, the picks-and-shovels layer is being built, and the tools for knowledge workers are becoming more capable.

All of that is accurate. It's also, from a CHRO's perspective, only half the picture.

The Productivity Layer vs. the Intelligence Layer

Claude Cowork is a productivity platform. That's not a criticism — productivity is genuinely valuable, and the HR workflows it targets are legitimately burdensome. The time an HR business partner spends composing a boilerplate offer letter is time not spent on the employee retention conversation that actually required human judgment. Automating the former to enable the latter is sensible.

But productivity platforms and people intelligence platforms answer different questions. A productivity platform answers: how do we do this work faster? A people intelligence platform answers: who in this organization is doing what, at what cost, with what dependencies, and what happens if they leave?

As Claude Cowork deploys across enterprise organizations, it will generate exactly the kind of footprint that people intelligence platforms are built to analyze. Every team that adopts an HR plugin, every manager whose workflow becomes dependent on an AI agent's output, every process that migrates from human execution to AI execution — that's organizational infrastructure being built. And like any infrastructure, it has an ownership question attached to it.

The Ownership Problem Is the CHRO's Problem

Here is the scenario that will play out at enterprise organizations in the next 18 months, at scale:

A senior HR business partner at a mid-size company is an early adopter of Claude Cowork. She builds a customized plugin workflow for the company's offer letter process — connecting Google Drive templates, DocuSign routing, and compensation band data. The workflow is sophisticated. It reflects institutional knowledge about the company's HRIS fields, its approval routing quirks, and the specific context that makes offer letters compliant with the company's total comp strategy.

Six months later, she leaves. The workflow breaks. IT doesn't know what she built or where the connections live. Her replacement knows the offer letter needs to go out but doesn't know the workflow exists, let alone how to reconstruct it. Three months of manual offer letter processing follow while the company figures out what they lost.

This is not a hypothetical. It's the calendar invite, the saved Slack workflow, the Excel macro, and the Gmail filter — all of which organizations have been losing when people leave for decades — except now those artifacts are AI agents with API integrations, data connections, and organizational dependencies that are harder to document and much harder to replace.

• Agent complexity scales with capability. The more powerful the plugin, the more institutional knowledge baked into its configuration — and the harder that knowledge is to transfer.
• Ownership is rarely explicit. Most enterprise AI deployments have a user-creator, not a designated organizational owner. When the user-creator leaves, the ownership question becomes acute.
• Dependencies are invisible until they break. Other workflows may depend on the output of an AI agent without anyone having documented that dependency. The failure mode is discovered, not anticipated.
• Cost concentration follows flight risk. The employees with the most sophisticated AI workflows are often your highest performers — the ones most likely to be recruited away.

What Claude Cowork Doesn't Know About Your Organization

Anthropic's enterprise pitch is clear: Claude Cowork gives every knowledge worker their own custom agent. That vision is directionally correct, and the execution is credible. The plugins are real, the integrations are with tools enterprises already use, and the HR workflow automations address genuine pain points.

What Claude Cowork doesn't — and by design, shouldn't — tell you:

• Which employees are running the highest-value AI workflows in your organization
• Which departments have the highest AI adoption concentration and which have none
• Which managers have teams with sophisticated AI infrastructure vs. manual workflows
• Which individuals represent single points of failure for AI-dependent processes
• Which employees running mission-critical AI workflows have elevated flight risk scores

This isn't Anthropic's problem to solve. It's the CHRO's problem to anticipate. The answer to it isn't a productivity platform — it's a people intelligence platform that treats AI agent ownership as an organizational variable alongside headcount, cost, and engagement.

The Market Reaction Tells a Partial Story

The software market's reaction to the Claude Cowork announcement — Salesforce up 4%, Thomson Reuters up 11%, the broader software ETF up over 1% — reflects investor confidence that enterprise AI automation is becoming infrastructure. That confidence is warranted.

What the market is pricing is the productivity layer: more work done, faster, by fewer people. What it isn't yet pricing is the organizational intelligence layer: the question of who controls the AI infrastructure, what happens when they leave, and how organizations will maintain visibility into their hybrid human-AI workforce at the moment it becomes genuinely complex to understand.

That question is the next category. It's not a feature of a productivity tool — it's a separate problem requiring separate infrastructure.

What CHROs Should Do Before Claude Cowork Scales

Three things matter before this adoption curve accelerates past the point where retroactive inventory is feasible:

Build an AI agent registry now, while the footprint is still manageable. Every AI integration, every workflow that touches organizational data, every plugin that routes work through an automated agent — document the owner, the business function, the dependencies, and the data access. This is straightforward when there are 12 workflows. It's a crisis management exercise when there are 1,200.

Map AI adoption concentration against flight risk. Your highest AI adopters are almost certainly also your most adaptable and marketable employees. The intersection of "runs our most critical AI workflows" and "has elevated flight risk signals" is where organizational single points of failure live.

Define what "AI agent ownership" means in your organization before a departure forces you to. Who inherits an AI workflow when its creator leaves? IT? The manager? A designated ops function? The answer doesn't matter as much as having one before it's needed.

Claude Cowork is genuinely good news for HR productivity. The offer letters will be better, faster, and more consistent. The onboarding materials will actually get made. The routine knowledge work that consumed HR capacity will increasingly not need to.

The question that follows is not about the tools. It's about what your organization looks like six months from now — which workflows exist, who owns them, which ones are mission-critical, and which of the people who built them are thinking about leaving. That question has always been the CHRO's question. In the AI era, it just acquired new urgency.

For how AI agent ownership maps to organizational risk when a key employee departs, see Your AI Agents Have Owners. For the broader thesis on why people intelligence can't be replaced by workflow automation, read The Real Intelligence Gap Isn't Artificial.