The Most Dangerous Thing About Your AI Coworker May Be Calling It a Coworker

The Most Dangerous Thing About Your AI Coworker May Be Calling It a Coworker

Ernesto Verdugo

"I Worked With Alice"

AI agents are entering companies as digital coworkers, and organizations don't yet know how humans should interact with them.

WIRED reports that AI agents are being given names, avatars, job-like roles, and even spots on corporate org charts. At companies using platforms such as Skydive, workers already say things like: "I worked with Alice."

The system is not being experienced psychologically as: "I used software."

It is becoming: "Alice helped me."

That's not semantics.

That changes supervision.

The Subtle Trap: Synthetic Authority

In BCG-linked research, when managers were told work had been produced by an "AI employee" rather than an "AI tool," they caught 18% fewer errors. The underlying research argues that treating agents as employees changes how humans evaluate and supervise them.

This is the core danger:

We may be changing human behavior around AI faster than we are changing AI capability.

The moment we call an AI a coworker, we may begin granting it social authority that its technical reliability has not earned.

I call this:

Synthetic Authority

Synthetic Authority is the unearned credibility, deference, or decision weight humans grant an AI because it has been presented with the social signals of an organizational role.

This is different from technical authority. A company might technically give an agent permission to read Slack. But psychologically, employees may give it much more: credibility, deference, assumed expertise, permission to direct work, and trust. Because it is called "Chief of Staff" instead of "Automation Process 47."

We May Be Manufacturing Authority Through Interface Design

Imagine the exact same output shown to a manager.

  • Version A: Generated by AI analysis tool
  • Version B: Recommendation from Sarah, your AI Strategy Director

Same model. Same answer. Different social framing.

The BCG finding suggests humans may inspect those outputs differently. We may be manufacturing authority through interface design.

The irony is that companies are anthropomorphizing agents deliberately because it makes adoption easier. WIRED quotes one AI coworker company saying personification helps because many people still don't understand what an agent is. Agents get names and Muppet-like avatars precisely to make their role intuitive.

So, the design decision that makes AI easier to adopt may also make it harder to supervise.

What Happens When Software Acquires Organizational Status?

An employee has: responsibilities, accountability, supervision, authority, disciplinary mechanisms, legal status, and performance reviews.

An AI "employee" has what exactly?

  • If Alice sends the wrong email, who gets disciplined?
  • If Alice makes a discriminatory recommendation, whose judgment was it?
  • If Alice promises something to a customer, did the company promise it?
  • If Alice tells a junior employee what to do, is she now a manager?
  • If Alice delegates to another agent, who hired that employee?

The Progression of Perception

The important change may happen even when the underlying model has not changed at all: the human relationship to its output changes.

  • Tool: "I use it."
  • Assistant: "It helps me."
  • Coworker: "We work together."
  • Expert: "I trust its judgment."
  • Manager: "I follow its direction."

The Workforce Is Already Changing

Millennium Management is expanding personalized AI "Digital Twins" from a pilot to more than 7,000 employees, while keeping them as assistants rather than decision-makers. PwC's new global workforce survey also found a widening AI divide: 14% of workers fall into its "front-runner" group, while 56% sit in the much larger "engine room" group with weaker access to AI and learning resources.

The emerging workforce problem is not simply humans versus AI. It is humans with radically different levels of AI augmentation, working alongside agents with radically different levels of authority.

We spent more than a century developing management systems for humans. Now companies are beginning to place nonhuman actors inside those systems before the management vocabulary has caught up.

The Recursum Connection

Recursum has been obsessing over the distinction that becomes critical here: Identity ≠ Access ≠ Authority.

  • Give Alice a name. Fine.
  • Give Alice an identity. Fine.
  • Give Alice access to Slack. Separate question.
  • Give Alice permission to approve something. Separate question.

Allow Alice to direct humans. Completely different question.

A job title is a social signal. It should never be mistaken for a permission system.

An org chart is not an authorization system.

Calling an AI "Chief of Staff" should not silently grant it the authority humans associate with a Chief of Staff.

A machine can have identity, organizational role, access, and technical permissions, while still having explicitly bounded authority.

The Most Dangerous Thing

The danger is not that AI becomes human. It is that humans start treating AI as though it were.

Because everyone expects: "AI coworkers are coming."

Very few expect: "Calling them coworkers may itself alter our judgment."

Ending

The most dangerous thing about your AI coworker may be calling it a coworker.

We are giving AI:

  1. A human name.
  2. A profile picture.
  3. A job title.
  4. A Slack account.
  5. An email address.
  6. A place on the org chart.

And in doing so, we are changing how we interact with it.

Not because its intelligence changed.

But because its social status changed.

We are manufacturing authority through interface design, subtly altering human scrutiny, and creating a new class of organizational actor whose responsibilities, accountability, and legal status are entirely undefined.

The workforce is transforming faster than our management vocabulary can keep up.

The question is no longer merely whether AI can do the job.

It is whether companies can distinguish the appearance of organizational authority from the authority the machine has actually earned.

Because once software has a name, a face, a title, and a place on the org chart, humans may start treating it differently long before the governance catches up.

The danger is not that AI becomes human.

It is that we forget it isn't.

TL;DR: The most dangerous thing about your AI coworker may be calling it a coworker. Companies are giving AI agents human names, avatars, job titles, and places on org charts, which leads humans to grant them "synthetic authority," social status not earned by technical reliability. BCG-linked research shows managers caught 18% fewer errors when work came from an "AI employee" versus an "AI tool," indicating human scrutiny changes with social framing.

This anthropomorphization, while making AI adoption easier, makes supervision harder. The article poses critical questions: Who is accountable if an AI "coworker" errs? What happens when software acquires organizational status before its implications are defined? The perception shifts from "using software" to "working with Alice," fundamentally changing human interaction and oversight.

The danger is not AI becoming human, but humans treating AI as if it were. The progression from Tool to Assistant to Coworker to Manager changes human relationship to output, not necessarily AI capability. Recursum emphasizes separating Identity, Access, and Authority. A job title is a social signal, not a permission system.

An org chart is not an authorization system. Companies are creating a mixed human-AI workforce without a clear management vocabulary. While companies like Millennium Management are scaling "Digital Twins" for thousands of employees and PwC notes a two-speed workforce emerging, the central issue is that calling AI a coworker subtly alters human judgment.

By Ernesto Verdugo, AI Architect, Founder of Verdugo Labs, and Creator of Recursum. He builds systems for what happens when human and artificial intelligence stop working separately.

Links: