As AI agents move from the factory floor into the office, the questions they raise are no longer purely technical. Who owns the output when a human and an AI produce something together? What happens to expertise when the work no longer requires people to understand it? And where should companies draw the line between what an AI can decide and what must stay with a named human being?
In conversation with Finbold, Terence Kwok, Founder of Humanity, addresses the emerging reality of AI as a workplace participant, from multi-agent coordination and digital identity to what it actually means to retire an agent responsibly.
Could an AI become your manager or manage other AI agents?
AI agents are already coordinating and supervising other agents in multi-agent systems. One agent can assign work and check whether another agent completed it. It can also escalate when something falls outside its permissions. To me, that is coordination rather than management in the human sense.
With people, the line should be much harder. An AI can help allocate work or identify a missed deadline. I would not want it to make final decisions about someone’s pay or promotion. Those decisions affect a person’s livelihood and need a named human who accepts responsibility for them.
The harder a decision is to reverse, the less comfortable I am with an AI making it on its own.
Who gets credit when a human + AI team produces something?
I think credit should follow who actually contributed what. OpenAI’s Navier–Stokes announcement is a good example. The company says an internal AI system produced a solution to a problem that had remained unresolved for roughly ninety years. Human researchers still chose which versions of the problem to test, directed resources toward the most promising route, and helped bring together the results from different agent groups.
Cases like this make it harder to attach authorship to a single name. We should be able to see who contributed the expertise, who directed the work, and what the AI produced. The same principle applies to responsibility. I still want a named person who can stand behind the final result.
If all of that gets reduced to the name attached to the finished work, we lose a useful record of how the result was actually created.
What happens to workplace culture when some “employees” don’t have consciousness, emotions, or even a physical presence?
Workplace automation began long before the current wave of AI. Siemens says machines and robots independently handle around 75 percent of the value chain at its Amberg plant. Rolls-Royce has described production as fully automated. Amazon has now deployed its one millionth robot.
What AI changes for me is where we encounter that automation. AI has moved beyond the factory floor and into the office. Software can now take part in work that used to depend much more heavily on human judgment and communication.
I see this as part of how work evolves. The harder question is where companies draw the line. An AI coworker does not get tired, worry about being excluded, or need recognition for good work. I would be concerned if companies started treating that as a standard for human employees.
Could an AI have its own digital identity, responsibilities, and reputation inside a company?
We have spent years giving AI systems a recognizable name and persona through products like Siri and Alexa. Now AI agents are starting to gain a formal digital identity inside companies. Microsoft Entra is already moving in that direction. An agent can have its own identity inside a company, and there can still be a person attached to it who is responsible for why it exists and how it is used.
I should be able to tell who gave the agent access, what it was allowed to do, and what it actually did. Its actions should remain traceable even when it acts without a person approving every individual step.
Reputation could work in much the same way. I would like to know how well the agent has done its job, how often people have had to step in, and whether it has caused problems before. That history should count before you give it more authority.
What happens to expertise when people no longer need to know how to do the work themselves?
I think expertise becomes more valuable. Aviation offers a useful parallel. Autopilot has handled routine parts of flying for years, but pilots still need to maintain their manual skills. Regulators keep stressing that too much reliance on automation can leave people less prepared to step in when something goes wrong.
I see the same risk with AI at work. A company can make people much faster without making them more knowledgeable. If nobody in the organization can challenge the answer, the company has outsourced more than execution. It has outsourced its ability to judge the work.
Will people eventually have personal AI agents that follow them from company to company?
They will, but the boundary has to be clear.
When someone changes jobs, they take their experience, working habits, professional relationships, and parts of what they remember with them. That has always been manageable because human memory cannot recall every conversation, document, or decision on demand.
An AI agent could take that much further. Its memory could be far more complete and searchable, and it could pull any of it back up instantly.
The agent should be able to keep the context that belongs to the individual, such as working preferences, skills, and professional history. Company records, confidential documents, and access rights should stay with the employer.
Could you have a rivalry with an AI coworker?
If people already compare themselves with their coworkers, why would AI be any different? The comparison has to feel fair, and companies should not judge people and agents as if they were doing the same job.
Recent research has found that employees can become less willing to collaborate with robots when they feel their status is under threat. Another experiment found lower performance in competitive teams with AI teammates than with human teammates.
I think the strongest rivalry will appear when companies treat people and agents as interchangeable units of output. An AI has no fatigue, career anxiety, or family obligations, so the comparison can become distorted very quickly.
I would look at what the person and the AI actually contributed together. If the AI handles 500 ordinary tickets and the employee takes on the ten difficult ones, the numbers by themselves do not tell you whose work was more valuable.
Best way to retire an AI agent?
I would treat retiring an AI agent much like offboarding a highly privileged account. Turning it off is only the beginning.
Access should end first. Its credentials, active tokens, and permissions should be revoked, and any connections to other systems should be closed. I would then want confirmation that the agent can no longer reach internal data or use any of the tools it previously had access to.
I would preserve its operating history. A company may later need to know which version was active, what it accessed, what actions it took, and who authorized it.
For me, an agent is retired when it has no remaining authority and the company still has a reliable record of what it did while it was active.
Read more interviews here.
The post Humanity founder Terence Kwok on what happens to work when AI joins the team appeared first on Finbold.