The Procurement Role That Turns AI Agents Into Infrastructure

There is probably already someone on your team who has become the default AI expert. It’s not in their job description, but they are the one everyone checks with when automating.

Deloitte found that 84% of companies have not redesigned jobs for AI. Expectations for what AI will automate keep climbing, and yet job descriptions have not changed. Luckily, procurement has a clearer place to start than most functions.

Why the AI skills gap is no longer the obstacle

Ask your team what is hard about working with AI Agents. In all likelihood, the answer you get today is different from the answer you would have gotten in January.

KPMG’s AI Pulse survey talks to the same population of enterprise leaders every quarter, and they reported some changes in how teams feel about AI Agents. Among the reasons employees push back on AI Agents, insufficient training and support dropped from 53% to 27% in a single quarter, skills gaps went from 76% to 57%, and job security fell from 67% to 55%.

Every fear that is on a standard AI change management plan is receding. People are getting comfortable with the tools, and they are getting comfortable fast.

One driver moved the opposite way. KPMG found that increased workload and complexity climbed from 28% to 51%. Procurement teams are no longer saying they lack the skills. They are saying the work arrived faster than the shape of their job changed.

Somebody has to tighten what the Agent knows, widen what it handles alone, and retire the checks that stopped earning their place. That is where the job description needs to catch up.

What an Agent Ops role owns

How many Agents are running in your procurement process right now, and who owns each one?

You might already have an operations person. Procurement Ops owns the platform, the workflows, the integrations, and the process documentation. That role has a counterpart now, but it's missing from most org charts.

Agent Ops is a role to own the Agents running inside the process. Where Procurement Ops asks whether a workflow is configured correctly, Agent Ops asks whether the Agent working inside that workflow produces output your team trusts, and how much it should handle on its own.

The work looks like this:

• Design: Build Agents for specific process steps, and give them the context to do the job well.
• Quality: Review output on a schedule rather than when you get complaints.
• Boundaries: Set the line between what an Agent completes independently and what routes to a person, and move that line as accuracy proves out.
• Escalation: Own the path for cases that fall outside what the Agent should decide.
• Retirement: Shut down Agents no one uses or ones that have been replaced with better versions.

Your Agent Ops hire should have three skill sets: procurement domain knowledge, technical fluency with AI platforms, and process design.

Domain knowledge is the one that is the easiest to pull off the job description, but it is the one that pays for itself fastest. Somebody reviewing a supplier risk score needs to recognize a good one on sight. That judgment turns review into a thirty second task instead of a thirty minute one, and it's what lets you hand the Agent more of the work next quarter.

That combination frequently already exists on your team. It tends to be the person who built your intake form, knows which categories the process actually breaks on, and gets all the Slack messages that say "can you look at this before I send it?" Giving that person the title and the hours is often faster than opening a search.

Why procurement needs Agent Ops now

Procurement is a function where Agents can touch spend, risk, and outside relationships. If a sourcing Agent scores a supplier badly and it costs your organization a contract, you are the one explaining it to the CFO.

That raises the value of the scope decision. What an Agent finishes on its own, in which categories, above and below which thresholds, is a procurement decision, not a technical one.

If that scope is not owned by anyone in particular, it freezes. Teams stall at the settings they configured on day one, and capability sitting in platforms goes unused.

Your first week with Agent Ops

You don't necessarily need a full-time hire to start creating Agent Ops. Start with these steps:

1. Count how many Agents are running. List every Agent live in your process today, and who reviews its output. Most teams find at least one that nobody owns.
2. Find your informal reviewer. Identify whoever keeps getting the "can you look at this automation before I push it live?" messages. That person is already doing the job.
3. Write down one boundary. Pick a single Agent and record what it is allowed to complete on its own. 
4. Set a review cadence. Put a recurring thirty minutes on the calendar to look at output quality, so issues surface on a schedule instead of in an emergency.
5. Give it a name. Add Agent Ops to the org chart, even at a fraction of an FTE. The title will make the eventual FTE defensible.

Somebody on your team is already doing this work. KPMG found that 47% of leaders say the people who have gotten good at AI are pulling ahead of colleagues who have not, and nearly eight in ten leaders would pay a premium of 6% to 15% for those skills.

That premium gets paid somewhere. It’s more efficient if you pay it to the person who already knows your categories.

We made the case for creating an Agent Ops role back in May. The data has caught up. The job is real, and the procurement teams getting serious impact from AI Agents are the ones who staff it.