AI agents enter company org charts and become coworkers

The agent became a "coworker": what BCG's research shows

In January, consulting firm Boston Consulting Group surveyed 1,261 managers and found that 22% of organizations had already added AI agents to the corporate org chart, not as a tool, but as if they were employees with a defined role. BCG partner Julie Bedard, who has been following the topic since 2023, says the lack of vocabulary to describe these systems is a symptom of a bigger problem: no one has yet decided whether an agent is a tool, a coworker, or something in between (Wired).

The distinction isn't semantic. When Microsoft's Copilot and Google's Gemini for Workspace hit the market in 2023, the pitch was "a tool that helps you work faster." What's changing now is that recent products are being sold as replacements for a role: in June, Microsoft launched Scout, an agent that reschedules meetings and drafts emails, with corporate vice president Omar Shahine describing the product as "your company hiring your assistant" that works "when you're not working."

From tool to digital employee: the Skydive case

The clearest example of an agent treated as a person is Skydive, a platform launched in August by vibe-coding startup Anything. Each Skydive agent has its own name, a fixed role, an avatar (Muppet-style, "close to human, but definitely not human," according to CEO Dhruv Amin), and its own cloud computer. One of the agents, named Canary, monitors production 24 hours a day. Amin justifies the personification: "most people still don't know what a real agent is," so giving it a name and a face makes adoption easier.

Pronto Housing CEO Christine Wendell is an early Skydive customer and uses agents for contracts, scheduling, and client follow-up. The detail that matters to those building software: Pronto's engineering team started using a coding agent named Alice, and Wendell noticed that engineers themselves began saying they "worked with Alice" on a project, the same verb they'd use for a human peer.

What changes in practice for those who review code

This slippage in language, treating the agent as a coworker instead of a tool's output, has a side effect measured by BCG itself: in controlled tests, managers identified 18% fewer errors when told the work had been done by an "AI employee" than when told it came from an "AI tool." The name given to the agent changes the level of scrutiny it receives, even when the output is identical.

For engineering squads, this is the most concrete takeaway from the report: if a coding agent like Alice shows up on a Jira board or signs a pull request under its own name, the risk isn't that the agent will make a mistake, it's that the human reviewer will relax their standards because they're dealing with a "coworker" instead of a script. Bedard sums up the state of the art at the companies she studies: "they're only just beginning to understand that guardrails apply here too." There is, today, no market standard for:

• An audit trail of what the agent decided on its own versus what a human approved
• The level of review for agent-generated PRs (is it the same checklist as a human PR, or a lighter one?)
• Who is accountable when the agent makes a mistake: the team that "hired" it, the platform vendor, or no one

These are process questions, not model questions, and they fall to whoever designs the CI/CD pipeline and code review policies, not to the AI vendor.

The counterpoint: using the name without creating a bond

Not every company that puts an agent on the org chart bets on personification. Lattice, an HR tech company, added digital employees to its own org chart two years before the topic went mainstream, but CEO Sarah Franklin is explicit: the goal is to reinforce accountability (knowing who did what), not to encourage the team to treat the agent as a person. Franklin sees the personification pushed by products like Skydive and Muse (a personal agent launched by Meta) as a retention strategy built on emotional attachment, and questions the underlying motive: "what's the purpose? Is it to push people out of work?"

This tension, between naming the agent to ease adoption and keeping clear that it's a reviewable system, is the practical point for engineering. An agent without a name and without an avatar remains easier to treat as what it is: an automated process that needs logs, versioning, and testing, just like any other system component. The choice to personify (give it a name, an avatar, and let the team say "I worked with Alice") is a product decision that platform teams and technical leadership should discuss before simply accepting a vendor's ready-made package.

What remains open

The Wired report doesn't provide adoption numbers for Brazil, and BCG's research covers managers in general, not engineering teams specifically. What can be taken from the source is the pattern: agents gaining names, roles, and a place in work tools (Slack, email, iMessage) at a faster pace than the pace at which companies are creating review and accountability policies for them. For those leading squads and deciding whether to buy a Copilot, a Scout, or build their own agent, the concrete lesson is to separate two decisions that today come bundled together: giving the agent a nice name to ease internal adoption, and defining the level of technical scrutiny that agent will receive, which shouldn't change just because of the name.