Agentic AI is no longer a future concept. It is arriving in enterprise environments right now, and with it comes a question that most organizations have not answered well: who is responsible for making sure the AI does the right thing?
The answer, more often than people realize, should involve a business analyst.
This is not a small observation. It represents a meaningful shift in how we think about the BA role in technology projects. For years, BAs were primarily responsible for capturing what humans needed and translating that into system requirements. In an agentic AI environment, the scope of that work expands in ways that are both challenging and genuinely important.
What Makes Agentic AI Different
Traditional software follows rules. A developer writes logic, the system executes it, and a human reviews the output before anything consequential happens. The loop is tight and the human control points are obvious.
Agentic AI works differently. An agentic system can perceive its environment, make decisions, take actions, and adapt its behavior based on what it encounters along the way. A well-designed agentic workflow might involve multiple AI agents passing work between each other, triggering processes, updating records, and completing tasks with minimal human involvement at each step.
That capability is powerful. It is also where things can go wrong in ways that are harder to anticipate and harder to catch.
When an agentic system takes an action based on flawed reasoning, incomplete context, or a misunderstood objective, the consequences can propagate quickly. By the time a human notices, the AI has already done significant work in the wrong direction.
This is not a hypothetical risk. It is a pattern organizations are encountering as they deploy agentic systems into real workflows. And it is precisely the kind of problem that a skilled business analyst is positioned to help prevent and manage.
The Three Levels of Human Oversight
There is a useful framework for thinking about how humans stay involved in agentic AI systems. It describes three positions a human can occupy relative to an automated workflow.
The first is Human-in-the-Loop. In this model, a human approves or reviews each significant AI action before it executes. The human is a required checkpoint. This provides strong control but can slow things down, and it is not always appropriate for every task.
The second is Human-on-the-Loop. Here, the AI proceeds independently, but a human monitors in real time and can intervene if something looks wrong. The human is watching, not approving. This is faster but requires the human to actually be watching and to have enough context to know when something is off.
The third is Human-out-of-the-Loop. The AI operates autonomously and humans review outcomes after the fact. This is appropriate for lower-stakes, well-defined tasks where errors are recoverable and the cost of human review at each step is not justified.
Organizations deploying agentic AI need to deliberately analyze and consider what steps and tasks require which level of human involvement and design the process and governance for the human role and process. Sitting down to decide which oversight model applies to which workflow and why is not an easy or quick conversation. A business analyst’s role is to facilitate this process and implementation with business, vendors, and engineering teams.
What the BA Actually Does in an Agentic Environment
The BA’s work in an agentic AI context is analytical before it is technical. It starts with questions.
What process and tasks are appropriate for agentic AI? What decisions is the AI making in this workflow? Which of those decisions are consequential enough that a human should approve them? Which ones can be monitored rather than pre-approved? Which ones carry enough risk that we should build in automatic escalation rather than assume someone is watching?
These are not questions a developer answers. They are not questions a project manager answers. They are business and process analysis questions, and answering them well requires someone who understands the business context, the stakeholder landscape, AI literacy, the regulatory environment, and the cost of getting things wrong.
Beyond oversight design, BAs also bring critical thinking to the knowledge structures that agentic systems rely on. An agentic AI is only as good as its understanding of the problem space it is operating in. When that understanding is incomplete or inaccurate, the AI will make decisions that seem logical from its perspective but are wrong from the organization’s perspective.
Building, validating, and maintaining those knowledge structures is analysis work. It requires someone who can elicit the right information from subject matter experts, identify gaps and contradictions, and translate messy human knowledge into something an AI system can use reliably.
BAs are also well positioned to design the exception handling that makes agentic systems resilient. Every automated workflow will encounter edge cases the original designers did not anticipate. What happens when the AI encounters a situation outside its training? Who gets notified? What is the fallback process? How does the organization learn from the exception and update the system?
These questions have always been part of good requirements work. In an agentic AI context, they become more consequential because the system is operating with more autonomy and the edge cases may surface in more unexpected ways.
The Risk of Skipping This Analysis
One pattern worth naming directly: many organizations are moving fast on agentic AI deployments and skipping the analysis work that would make those deployments safer and more effective.
The motivation is understandable. Agentic AI tools are compelling, the competitive pressure to deploy is real, and analysis feels like it slows things down. But the shortcuts tend to surface as problems at the worst possible time, when the system is in production, handling real decisions, and the consequences are visible to customers or leadership.
I like to think of this as discipline, rather than a phase, or slowing things down.
The analysis work is not a gate that delays deployment. It is the work that makes deployment sustainable. Organizations that get this right spend more time on design and less time on remediation. They build systems that can be trusted, monitored, and improved over time rather than systems that require constant firefighting.
These systems need continuous monitoring, and they need incremental and iterative development. Our funding and work methods need to be adaptive and incorporate learning along the way.
Building This Capability
If you are a BA who wants to be effective in agentic AI environments, there are specific capabilities worth developing.
Start with AI fluency. You do not need to build AI systems yourself, but you do need to understand how they work well enough to ask the right questions and identify the right risks. This means understanding what large language models are good at and where they fail, what agentic patterns look like in practice, and how multi-agent systems are typically structured.
Add process analysis depth. Agentic AI changes processes in ways that are not always obvious. A workflow that previously required five human decisions might now require one, but that one needs to be designed carefully. Getting good at mapping agentic workflows, identifying decision points, and designing oversight structures is the core skill.
Develop your risk vocabulary. Being able to articulate the risk profile of an agentic deployment in terms that resonate with leadership is increasingly important. BAs who can frame the trade-offs clearly will be more influential in the decisions that matter.
These capabilities are teachable and learnable. They do not require a technical background. They require the willingness to engage with new concepts and apply the analytical thinking BAs already do to a new category of problems.
If you want to go deep on all of this, I teach it in my Maven AI/BA course series. We cover the analysis behind enterprise agentic AI, human oversight models, and the practical skills BAs need to work confidently in these environments. You can find me at maven.com/angela-wick.
The Bigger Picture
The organizations that deploy agentic AI well will do so because someone on their team was asking the right questions before anything went into production. Someone mapped the decisions the AI would make and designed the right level of human oversight for each one. Someone built the knowledge structures that let the AI understand its context. Someone designed the exception handling that kept things from going sideways when the system encountered something unexpected.
That someone is, very often, a business analyst.
The role is not shrinking in an AI world. It is migrating toward exactly the kind of strategic, analytical, human-centered work that AI systems cannot do for themselves. The question for every BA right now is whether they are developing the skills to occupy that space.