Anthropic, one of the most prominent developers of agentic AI systems, recently posted a job listing for a human executive assistant in its London office, offering a salary of approximately 145,000 euros, or roughly $190,000. The role was described as a position for a “highly organized, detail-oriented, and self-starting administrative business partner.” The listing has since been removed from Anthropic’s website. What makes this story worth examining is not the salary itself, but what the hiring decision reveals about the gap between what agentic AI can do in theory and what organizations actually trust it to do in practice.
What the Role Actually Required
The job description asked for someone to handle scheduling, travel logistics, inbox management, expense filing, and general executive support for senior go-to-market leaders. These are not exotic tasks. They are precisely the kind of structured, repetitive, coordination-heavy workflows that agentic AI systems are designed to automate.
Anthropic’s own products are built around the idea of AI agents that can take sequences of actions autonomously, navigate digital environments, and complete multi-step tasks without constant human supervision. The company is not a peripheral player in this space. It is one of the leading developers of this technology. That context is what gives the job posting its particular weight.
The listing specified requirements including “outstanding attention to detail, strong relationship-building skills, and the ability to anticipate needs across teams while ensuring responsive, error-free execution.” Each of those phrases maps directly onto capabilities that agentic AI systems are currently marketed as possessing, or approaching. And yet, for this specific internal role, Anthropic chose to hire a person.
The Irony That Social Media Noticed
The posting attracted attention online, with users pointing to the apparent contradiction between Anthropic’s public positioning and its internal hiring choices. One comment, widely circulated, suggested that someone should build a technology tool capable of handling these tasks autonomously. The joke landed because the punchline was obvious: Anthropic is, by most accounts, already trying to build exactly that.
This kind of public reaction is worth taking seriously, not as mockery, but as a signal. When a company’s external messaging and internal behavior diverge visibly, observers notice. The gap becomes a data point about where the technology actually stands, regardless of where the marketing says it is heading.
It is worth being precise here. The source does not indicate that Anthropic made any public statement explaining the decision, and Fast Company’s request for comment went unanswered at the time of reporting. The reasons behind the hiring choice remain unknown. What is known is the choice itself.
What This Gap Actually Tells Us About AI Adoption
Here is what most coverage of this story misses: the Anthropic case is not an anomaly. It is a visible instance of a pattern that plays out across the technology sector and beyond.
Organizations developing AI tools for external use routinely maintain human-staffed operations for the same functions internally. This is not necessarily hypocrisy. It reflects something more structurally interesting: the difference between a technology being capable of performing a task and an organization being ready to trust that technology with consequential, real-time operations that affect senior leadership.
Executive support is a high-stakes function. Errors in scheduling, miscommunications in inbox management, or failures in travel coordination have immediate, visible consequences for the people being supported. The tolerance for error is low. The need for contextual judgment, discretion, and relationship awareness is high. These are precisely the dimensions where agentic AI systems remain least reliable, not because the underlying models lack capability in a narrow sense, but because the integration of that capability into a live, accountable workflow introduces failure modes that are difficult to predict and costly to experience.
This is the distinction that matters. AI systems can handle these tasks in controlled demonstrations. Deploying them in a live environment, where the cost of a mistake falls on a senior executive and the organization’s reputation, is a different calculation entirely. Anthropic, a company with deep technical knowledge of what its own systems can and cannot do, apparently concluded that a human was the right choice for this role. That conclusion carries more information than any benchmark result.
The broader implication is that the adoption of agentic AI in workplace settings will not follow a clean, linear path from announcement to deployment. Organizations will continue to hire humans for roles that AI could theoretically fill, not out of ignorance, but out of a rational assessment of risk, accountability, and the current limits of trust in autonomous systems. The technology and the organizational readiness to use it are developing on different timelines.
In Short
Anthropic posted a human executive assistant role in London at a salary of roughly $190,000, for tasks that agentic AI systems are theoretically designed to handle. The company did not publicly explain the decision. What the episode illustrates is a gap that exists across the industry: the distance between what AI can do in a demonstration and what organizations are prepared to rely on it for in high-stakes, real-time operations. That gap is not primarily a technical problem. It is a trust and accountability problem, and it will shape AI adoption in the workplace long after the underlying technology has advanced further.
Based on reporting from Fast Company.