Where AI Agents Belong in Your Company (And Where They Don’t)

The AI we have access to today looks very different from what most of us were using just a few years ago. Early on, a lot of us treated AI like a more capable search engine or assistant. We’d ask a question and wait for a response.

But technology is moving so fast and agents today can do far more than answer a simple question. They can monitor information over time, alert us when something is off, help us evaluate a decision from several angles, coordinate across systems, and keep a process moving without waiting for a person to direct every step.

In the first article in my AI Leadership series where I explore how to bring AI into your company the right way, I focused on how founders can lead AI adoption thoughtfully. Now, I want to go one step further and look at where agents belong inside that picture. As the capabilities of AI agents expand, we’re looking for more than help answering questions. We’re beginning to rely on AI to help us run parts of our company better. And that raises a much bigger question for founders and leaders: 

What work should we delegate to an AI agent, and what should stay firmly in human hands?

I don’t think we can answer that by focusing on what AI is capable of today. Those capabilities are only going to keep advancing, and our answers will keep evolving along with them.

A better place to start is with how your company operates. Which parts of a process follow a clear pattern? Where does the work tend to slow down? Which parts depend on human judgment, context, or experience?

Those questions give you a much stronger foundation for deciding where agents belong and where they don’t. Let’s walk through how to determine what work you can delegate to an AI agent while keeping human judgment and accountability connected to every outcome.

Start With Your Processes, Not the Agent

When a new AI capability shows up, the first question is often, “Where can we use this?” But I think there’s a better place to begin: with the recurring journeys and processes already happening inside your company. Which processes already follow a clear, repeatable pattern? Those are the places worth examining first.

For example, maybe one of your leaders reviews data every Monday and identifies anything that needs attention. From there, they pull together relevant context and follow up with the right person. An agent may be able to handle several parts of that process before the leader ever gets involved. AI could monitor the data, flag anything that falls outside a set range, bring together the information needed to understand what’s happening, and route it to the right person for follow-up.

That’s where agents become truly valuable. When they take on parts of a repeatable process, your people have more capacity to focus on the decisions that require their judgment. That’s one advantage of using a business operating system like Ninety, where your priorities, roles, agreements, and performance information are already documented and living in one place. With AI built into the platform, the agent has a much clearer picture of how your company actually works.

This doesn't mean every recurring process should be handed to an agent. Before you delegate any part of a process to AI, you should be able to explain how that process is supposed to run, what information the agent would need, who is responsible for the agent (yes, they need to be given a name and added to your Org Chart), and when and how that person still needs to be involved. If you can’t explain those pieces clearly, the process probably needs a better structure before you bring an agent into it.

Keep in mind, the decision to stand up a new agent shouldn’t focus only on whether an agent can perform part of a process. You should also determine whether delegating part of the process to AI gives your team more capacity while still producing an outcome you can trust. Otherwise, you risk complicating something that’s already working well as it is.

Evaluate the Consequences of Getting It Wrong

Once you’ve found a process that seems like a good candidate for an agent, the next thing I’d look at is what happens if the agent gets something wrong (because AI does still get things wrong)?

Some mistakes are easy to correct. Others can create consequences that are much harder to undo. If an agent organizes an internal report incorrectly, someone can fix it and move on. But if AI sends the wrong email to an important customer, changes a financial commitment, or makes a decision that affects someone’s role, the stakes are very different.

That doesn’t mean AI has no place in higher-stakes situations. It may still help gather information, weigh options, or recommend a next step. But I’d be much more careful about how far you allow the agent to go on its own.

As you evaluate a process, ask:

• If the agent gets this wrong, how easy is it to correct?

• Could the outcome affect a customer, employee, commitment, or important relationship?

• Who will be accountable for reviewing the result?

• Can I clearly explain why this step is appropriate to delegate to an agent?

• Does the agent's manager have the Competency, Commitment, and Capacity to manage it? 

Those questions help you decide how much of a process you can delegate to an agent, if any. In some cases, the agent may be able to handle nearly everything. In others, you may find it’s not the right fit.

Either way, someone still needs to be accountable for the outcome. AI can take on parts of a process, but it can’t take accountability off someone’s shoulders. A person still needs to understand what the agent is doing, know the standard it’s expected to meet, and own the result.

That’s the balance we need to weigh as founders. The more autonomy you give an agent, the more important it is to understand the risk if it gets something wrong.

Decide How Far AI Should Go

Human judgment should stay involved anywhere AI is being used. The question is how much of a process you’re comfortable delegating before a person needs to step in. You can start by looking at what the outcome depends on.

Some processes may be structured and repeatable, but the final decision still relies heavily on context, relationships, experience, and values. Take a decision about whether someone is still the right fit for a Seat. AI can help you review and organize their performance history, summarize feedback, compare results with agreements, and bring useful context together. That can make the decision easier to prepare for, but it’s not a decision an agent should ever make on its own.

In a situation like this one, there are several factors you have to weigh that go well beyond what shows up in a report. How has this person developed over time? What conversations have already happened around their performance? What does the company need from this Seat going forward?

That same thinking also applies across other parts of the company. An agent might help prepare for an important customer conversation, but a person is still responsible for taking the meeting and keeping the relationship. Before a strategic decision, AI can help compare options or weigh risks, but the final choice must stay with the person or team. It can also analyze company performance and suggest where to adjust, but the people who understand the brand and broader strategy should make the final call.

As more capabilities become available, we have to be thoughtful about where the agent stops and where a person steps in. Human judgment still needs to guide the process, and someone always needs to be accountable for the outcome.

AI is a tool. The choice about how it gets deployed is ours.

Oren Etzioni

Founding CEO of the Allen Institute for Artificial Intelligence

How to Decide Where to Stand Up an Agent

Before you stand up an agent, you should be clear on what you want it to improve, whether that’s increasing efficiency, reducing repetitive steps, or giving your team more capacity for higher-level work.

I’d start with processes you already understand well, where the information is dependable, the boundaries are clear, and one person is accountable for the outcome. You also need to understand the risk if the agent gets something wrong. Together, those factors give you a much stronger basis for deciding whether a process is ready for an agent.

Before standing up any agent anywhere in your company, ask these seven questions:

• Is the process clear and repeatable? You should be able to explain how it runs today without filling in gaps as you go. The more consistent the process is, the easier it is to see where an agent can fit without creating confusion or unnecessary complexity.

• Is it clear who will be accountable for the agent? As the saying goes, "If two people are accountable, no one is accountable." Do not stand up an agent who reports to more than one person. 

• Does the agent have the information it needs? If important context lives only inside someone’s head, you probably aren’t ready to delegate that part yet. You can end up losing time if people constantly have to fill in missing information or correct the context before the agent can do its part well.

• Can you define where the agent stops? Be clear about where human review is still required, especially when the output could affect a customer, employee, or important business decision. Without clear boundaries, it’s easy for people to put too much trust in the agent and rely on it more than they should.

• Can you recover if it gets something wrong? The harder an outcome is to correct, the more closely a person should stay involved. You want to understand the downside before you give an agent too much autonomy.

• Will delegating this give your people meaningful capacity back? If the agent adds complexity without helping the company operate better, there may not be much reason to introduce it. When used well, AI should create more capacity for higher-value work, not add another system your team has to manage.

• Is the cost/benefit obvious? Agents add complexity to your organization, so the return should be extraordinarily clear before you introduce one. Right now, I’d want to see a cost-benefit ratio of at least 1:5.

Answering those questions gives you a clearer way to think about where agents can add value. We’re all going to learn as these capabilities develop. But you should be able to explain why you’re standing up the agent, what you expect it to handle, where its boundaries are, and who remains accountable for the outcome. If you can do that, you have a strong place to start. If you can’t, you may need to spend more time improving the process before adding an agent to it.

Keep Learning From What You Delegate

The first few agents you stand up will teach you a lot about your company. They’ll show you where your processes are stronger (or weaker) than you thought, where people are relying on context that lives in their heads, and where your agreements need to be clearer. In that sense, agents can become another way to learn how well your company is designed.

One of the most important parts of bringing an agent in is being clear about who owns the outcome. An agent can perform work on behalf of the company, but someone still has to be accountable for what it does and the consequences that follow.

That’s what I’ll cover in the next article in the AI Leadership series: who’s accountable for your AI agents, why every agent needs a human owner, and what that accountability should look like as agents take on more responsibility.

We’re all still learning what great AI use looks like inside a company. But that’s part of the opportunity in front of us. We get to shape how this technology fits into the companies we’re building. We get to decide what we’re willing to delegate and what should remain distinctly human. That’s an important leadership choice, and I think it’s one we’re going to be making for a very long time.