Boss, Coach or Assistant? The 7 Roles AI Can Play at Work

There's one question almost no team asks before bringing AI into its work: what role will it play? The tool gets chosen, the pilot gets launched, the speed gets measured. Meanwhile, the relationship between people and AI is left to each user's habits.

At the keynote of the 2026 Professional Development and Employment Fair (JIP 2026) at the Universidad Nacional de Córdoba, our Partner & Managing Director, Walter Abrigo, put that question to students about to enter the workforce. His answer was a framework that brings more clarity to the conversation than any list of tools.

Seven Ways to Work With AI

Deloitte's Global Human Capital Trends 2026 report describes seven types of interaction between people and AI, ranked from the most to the least authority AI holds over the person:

No relationship is better than another in the abstract. A coach helps you learn, an assistant handles one-off tasks, and an autonomous worker takes on repetitive, well-defined processes. Trouble starts when nobody makes the choice.

Only 14% of Leaders Know How to Design Them

According to the same report, just 14% of leaders surveyed say they're skilled at shaping these interactions. Everyone else lets them take shape on their own.

That has real consequences. An AI meant to be an assistant ends up making decisions no one reviews. Another that could run an entire process gets reduced to a chat window for occasional questions. Using AI is one thing; designing how we work with it is another.

Walter drew on his own experience as a professor, after automating part of his day-to-day work: "Some things can be automated, others need human intervention within the automated process, and others must, without exception, go through a person."

One Relationship per Task, Not per Company

The unit of analysis isn't the role or the company. It's the task. A single process can call for three different relationships:

• What can be automated works with AI acting as a direct report or an autonomous worker.

• What needs review calls for AI that prepares and a person who validates before moving forward.

• What has to go through a person keeps AI as a coach or collaborator, never as the one who decides.

Choosing well takes deep knowledge of the work itself. That's what lets you tell a useful answer from a wrong one and explain why a decision deserves trust. Without it, any relationship with AI becomes fragile.

What We Learned Running Our Own Agents

Before offering agents to clients, we put them to work in-house. Today Santex runs more than 20 of its own agents in production, every day, on Slack. Each one has a clear relationship with the team that uses it. Defining that relationship is the first step in designing conversational agents that teams actually adopt. 

That experience shaped STX Agents, our agentic AI solution for business operations. The agents read context, query systems and run processes, always within business rules, defined roles, full traceability and human approval wherever a step requires it. The relationship is decided before the first line of the agent is written, not after.

The path starts clear and concrete: one process, one agent, and metrics from day one. In three weeks, the agent goes from a discovery session to running in Slack, Teams or wherever your team already works.

Three Questions Before Adding an Agent to Your Company

1. Which tasks in this process can be automated, which need review, and which have to go through a person?

2. Who validates what the AI does, at what point and with what information?

3. How will we know the relationship is working, and when should we change it?

If your team already uses AI but hasn't decided what role it plays, that's the place to start. Book a 30-minute discovery session with us to choose the process and the relationship worth starting with.