In the Age of AI Agents, the Most Important Role Isn't Technical

Imagine you hire 100 new employees. They work around the clock. They never ask for time off. They can serve thousands of customers simultaneously, draft personalized emails, and qualify prospects without breaking a sweat. But they also make decisions on their own. They talk to your customers without asking permission. And when they don't know something, they sometimes make up the answer instead of asking.

Who do you put in charge of them? The CTO, who understands how they work but doesn't know your customers? The head of sales, who knows the business but has never managed anything non-human?

Those 100 employees are AI agents. And the question of who manages them is no longer hypothetical.

A role that didn't exist two years ago

Salesforce now operates dozens of AI agents on its Agentforce platform. They handle customer inquiries, draft emails, qualify leads, and escalate difficult cases to human specialists. They autonomously resolve roughly 74 percent of incoming support cases.

For that to work, the company had to create a role that didn't exist: the agent manager.

Zach Stauber holds that position at Salesforce. His daily routine involves reviewing dashboards, spotting where an agent responded with the wrong tone, investigating why another escalated a case it could have resolved on its own, and adjusting instructions so they do better tomorrow. "Data, data, data. I start and end my day in dashboards, scorecards, and agent observability monitoring," says Stauber. It sounds like any operations-obsessed manager. The difference is that none of his direct reports are human.

Stauber isn't an engineer or a data scientist. He studied audio production. He spent years in conversational design and led chatbot teams before landing this role. And that's no coincidence — it's the norm. The most effective agent managers don't come from technical teams. They come from operations, customer service, and sales. People who already knew what a good outcome looks like in practice.

In Latin America, this role doesn't exist. The conversation about AI agents in the region still revolves around which tool to buy, which platform to implement, and which vendor to hire. It's a conversation about acquisition, not operations. And the competitive advantage won't go to whoever adopts agents first, but to whoever learns to operate them first.

The business leads, not IT

For a long time, AI was an IT department matter. The tech team built it, deployed it, and maintained it. The rest of the company received the finished product and dealt with the results as best they could. That worked when AI was something passive — a predictive model here, a spam filter there.

With autonomous agents, that model breaks down. An agent isn't software that someone opens and closes. It makes decisions, talks to customers, and executes tasks without direct supervision. If nobody from the business side takes ownership, it will make mistakes that the engineering team can't even anticipate because they lack the context to understand them.

Salesforce figured this out quickly. Customer success teams — not engineers — are the ones who define the agent's tone, its escalation rules, and its success metrics. If the agent is doing real work for your department, you're responsible for making sure it does it well. That's why the most important role in this new era isn't technical. It's management, with deep business knowledge.

Companies that centralize agent management in IT will end up with agents that work technically but don't understand the business. And an agent that doesn't understand the business is like a salesperson who has the CRM memorized but can't read a customer.

From 150 meetings a month to 350 in a week

Consider what happened with Salesforce's sales development team. Before agents, their reps contacted dozens of prospects daily but only managed to speak with 12 to 15. The rest of the leads went cold because there weren't enough hands.

Today an AI agent handles the first interaction at scale: personalized outreach, qualification, and consistent follow-up with leads that no one used to touch. "While my team sleeps, our agents are already engaging with customers," says Vanessa Tabbert, VP of sales transformation and development at Salesforce.

The results showed up fast. They went from 150 meetings booked in 30 days to over 350 in a single week, with the same lead volume. In four months they generated $60 million in annualized pipeline and added more than 300 new customers. The team that pulled it off was small — what Amazon calls a "two-pizza team" — and quickly deployed the agent across the United States, Canada, the UK, Ireland, Africa, and Japan.

Those numbers didn't come from plugging in an agent and hoping for the best. Behind them was someone deciding when the agent should hand off to a human. Someone reviewing whether the instructions it received were producing the right behavior. Someone measuring whether the meetings it generated were turning into real business or just noise. Someone from the business, not from engineering.

Without that management layer, the agent would have been nothing more than a meeting-scheduling machine with no real value.

Operators with judgment, not engineers

What does an agent manager actually need? An understanding of how prompts work and how to diagnose failures. Deep knowledge of the business process the agent supports. Systems thinking — how agents interact with each other, with other teams, and with humans. The ability to adapt quickly when models or priorities change. The skill to write clear instructions that shape the agent's behavior, which is the equivalent of training a new employee. And the ability to design workflows where humans and machines complement each other without getting in each other's way.

Organizations that built these teams with people from operations and customer service — people with business judgment — scaled faster. Those that sought technical profiles or centralized everything in IT ended up with managers who knew how to pull levers in the system but didn't understand which levers mattered. What made Stauber a good candidate wasn't a graduate degree in machine learning, but what he describes as "genuine curiosity" — the drive to experiment, learn on the go, and take ownership of the outcome.

In the region, these management skills already exist — in operations managers, customer service leaders, and commercial directors who know their market intimately. What doesn't exist yet is the organizational decision to redirect those people toward agent management instead of continuing to delegate AI to the technology department.

Goodbye to activity KPIs

Managing agents also forces a rethink of how we measure people. Classic KPIs — 60 calls a day, 40 emails, 20 tickets — lose their meaning when an agent can do all of that continuously and at scale.

The Salesforce sales reps case illustrates this well. Before, a good salesperson was the one who made the most calls. Today, the agent makes the calls. A salesperson who still measures their value by contact volume is competing against a machine that doesn't sleep and doesn't get distracted. It's a race they can't win.

The metrics that matter now are different. How well is this salesperson fine-tuning their agent? Are they capitalizing on the meetings the agent books, or wasting them? Are they closing the deals that require empathy and human judgment? What percentage of agent-generated opportunities turn into actual revenue?

The focus shifts from counting individual activities to measuring the efficiency of the complete system — person plus agent working together.

For human teams, that means learning two things. The first is managing the agent: giving it clear instructions, reviewing its work, and adjusting its behavior when results aren't as expected. The second is being effective in the interactions the agent can't handle — complex negotiations, tough objections, moments where a customer needs to talk to someone who understands their context. Once again, business skills, not technology skills.

Back to those 100 employees

Agent management isn't a niche role. As agents move into finance, human resources, supply chain, and operations, every area of the company will need someone to take charge of them. In a year or eighteen months, "agent manager" will be a title as common as "product manager" is today at tech companies.

But the title matters less than the capability. And this is where the gap between markets will show.

In the United States and Europe, companies like Salesforce, JPMorgan Chase, and Walmart are already building dedicated teams to operate agents. They're defining metrics, creating continuous improvement processes, and distributing responsibility across every business unit. In Latin America, most companies are still in the buying phase: evaluating platforms, negotiating licenses, running pilots that live inside IT.

The difference between these two groups isn't budget or access to technology. The technology is the same for everyone. The difference is that some are already learning to operate while others still believe that buying is enough.

Let's go back to the opening question. You have 100 new employees who work nonstop, make decisions, and talk to your customers. They're fast, cheap, and scalable. But nobody told them how to handle an upset customer in Monterrey, or when it's better to pass the call to a human, or which metrics define whether they're doing a good job.

Who's in charge of that at your company? If the answer is "the IT team" or "nobody," you know where to start. And the right person for the role probably already works with you — they're just managing humans today.