Alex Saric is the chief marketing officer at Ivalua.gettyNearly every enterprise leader is somewhere on the same road right now. When AI moved from experiment to board-level priority almost overnight, the pressure to show results pushed most organizations into the same pattern: find a use case, deploy an agent, move to the next one.Having spent the last few years experimenting with LLMs, custom models and specialized agents at Ivalua, we’ve watched this arc play out. Each stage felt like progress, but each one also exposed something the previous stage missed.​​An important lesson is that more agents mean more to manage, and no one can supervise a hundred agentic specialists at once. While many leaders try to solve this problem by adding more AI, a better option is often giving every team member one expert: an agent that knows your data, works within your policies and answers to you. ​More Agents, More ComplexityThe appeal of multiple agents is obvious. Procurement alone has hundreds of distinct tasks—supplier onboarding, risk monitoring, contract review, spend classification and invoice validation, just to name a few. Build a specialist for each one, and you’ve covered the function. That logic drove most software vendors toward a similar architecture, but the problem with it showed up at scale. When agents multiply, so does the management burden. Someone has to watch what each one is doing, manage its permissions, verify its outputs, reconcile conflicts and audit the trail when something goes wrong. The supervision problem runs deeper than coordination. An agent handling supplier risk operates on different data than the one managing contract renewals—even when both concern the same supplier. Since they don’t share context, the person nominally overseeing both rarely has bandwidth to bridge that gap. You end up with capable AI and incomplete intelligence. According to a March 2026 survey of 800 procurement decision-makers across the U.S., U.K., Germany and France conducted by Sapio Research on behalf of Ivalua, 59% of procurement leaders say trust in AI is a bigger barrier to progress than cost or technical capability. The architecture itself is contributing to the skepticism.​The Wrong QuestionRather than asking how many agents you need, I recommend asking a harder question. How can AI support a better way of working? Your team becomes more effective as employees learn the organization and develop more skills, not by you hiring more employees with only specific skills.When AI acts, who is responsible? In a fleet of specialists, the answer gets murky quickly. Each agent handles its slice of the process, passes outputs to the next and the chain of accountability dissolves somewhere in the middle. Worse, each agent inherits the fragmentation of the data beneath it—supplier records in one system, contracts in another, spend history somewhere else. None of them work from a shared picture of the truth. When something goes wrong, you struggle to trace the decision and you can’t trust what it’s based on.The shift is to think of AI as part of your team and put the human back at the center of the new structure—not as a supervisor of many agents, but as the accountable owner of one. If one agent inherits each user’s permissions, works from their data, follows their organization’s established practices and policies and pauses for their judgment when the stakes require it, it becomes an expert in your business that knows when to ask for permission before continuing.Governed By DesignThe most visible sign that something fundamental has changed isn’t in the architecture but in how people work. When AI operates within a user’s specific permissions and workflows, the interface changes. A category manager can ask for a contract in context rather than searching for it, and a risk analyst can describe what they need instead of pulling a report. This is the true goal of agentic AI: the conversation becomes the interface.Getting there requires three things to be true simultaneously:• The AI has to be governed by design—not rules bolted on after the fact, but permissions inherited directly from the people it works with. It cannot access data for someone who isn’t authorized to see it or take any actions that exceed an individual employee's boundaries. Every action should be logged and attributable. When the process calls for human judgment, the agent must wait. • The expertise has to compound. The knowledge your best people carry—how to handle a specific category and when to escalate—should be captured as skills the AI leverages, so a new hire operates with the benefit of the whole organization’s experience from Day 1. • It has to fit how your organization already works, connecting to your existing systems, chosen models and established workflows without becoming another integration project.When all three are in place, people direct the AI, and the AI delivers. That’s the working relationship worth building toward.For procurement and supply chain leaders, the pressure is real and immediate. Tariff volatility, supplier risk and regulatory complexity are problems that must be solved today. In my experience, teams working with AI within governed boundaries, drawing on a single source of truth and compounding institutional knowledge over time, place themselves in a better position to respond faster and make better decisions than those still managing a roster of narrow specialists.​What Comes NextEvery genuinely transformative technology delivers an obvious win first. The printing press made reproduction faster. Early computers processed calculations at speeds no human could match. However, while those were real gains, they weren’t what made the technologies transformative. The lasting impact came when they changed who could access knowledge, how organizations made decisions and what became possible as a result.Agentic AI is in that same moment now. Most organizations are still capturing the obvious wins—automating tasks, researching more quickly or accelerating individual processes. The more valuable question is what it changes about how people work, who is accountable and how institutional knowledge gets built and preserved. That’s the question that led us here. Whatever comes next—and the technology will keep moving—the organizations that ask it early will be ahead of the curve.​Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?