Imagine hiring a sous-chef who cannot actually taste the food, cannot hold a knife, and occasionally insists that salt is a type of sugar. You wouldn’t call that person a peer or a collaborator; you would call them a liability. Yet, that is exactly the kind of semantic gymnastics currently being deployed by the corporate AI machine to make automation feel like a team-building exercise.
The attempt to rebrand agents as “coworkers” is a blatant psychological trick. As noted in MIT Tech Review, the industry is pivoting toward a model where these tools are integrated into the organizational chart. This isn’t about fostering a spirit of collaboration. It is about shifting the cognitive load and the blame. When a human coworker messes up a report, there is a performance review and a conversation about expectations. When an AI “coworker” hallucinates a set of Q3 projections, the manager gets to blame the “tooling” while still reaping the rewards of the reduced headcount. It is essentially gaslighting the remaining human staff into believing they are managing a team rather than babysitting a stochastic parrot. (And let’s be honest, most middle managers love the idea of having “direct reports” on their LinkedIn, even if those reports are just API calls).
Then there is the actual friction of the user experience (I suspect most “coworker” advocates have never actually used an agent for a complex, multi-step task). In the real world, this looks like agonizing API latency and the creeping cost of tokens that make a simple automated workflow more expensive than a junior intern. There is nothing “peer-like” about a coworker who takes four seconds to respond to a prompt and then requires three follow-up corrections because it forgot the primary constraint of the task. It is like working with a colleague who is incredibly confident but suffers from short-term memory loss every time the conversation hits a certain token limit. We have seen this movie before—remember when “no-code” platforms promised to replace developers only to end up as expensive toys for product managers? This is the same play, just with a different set of buzzwords.
The solar-powered internet angle is a curious addition to this narrative. The idea is to provide the connectivity needed to keep these agents running in the most remote corners of the globe. But why are we rushing to build the infrastructure to deliver these agents to rural areas? If the goal is truly productivity, why prioritize the connectivity for a bot over the connectivity for the person operating it? It feels like a hardware play disguised as philanthropy. We are essentially building a global nervous system for agents, ensuring that the “coworker” can report to the home office from a village in the Andes without a flicker in the connection, regardless of whether the human in that village has a stable enough connection to actually use the tool. Or maybe it’s just a way to ensure that the compute load is distributed across regions with cheaper energy. Either way, the focus is on the machine’s availability, not the human’s utility.
The industry is betting that if they change the name, we will forget the limitations. They want us to stop seeing the agent as a script and start seeing it as a colleague. But the gap between a sequence of probability weights and a professional peer is too wide to bridge with a marketing pivot. You cannot “culture” your way out of the fact that an agent cannot take ownership of a project or feel the heat of a deadline. By Q4 2026, the “coworker” terminology will have been replaced by “agentic workflow” in the majority of enterprise software marketing because the metaphor will simply collapse under the weight of its own inaccuracy. When the first few major corporate disasters are blamed on a “coworker” who was actually just a poorly tuned prompt, the PR department will pivot back to calling them “tools” real quick.
AI agents are just fancy scripts with better PR.
