Archive | Upgrade to Premium | Sponsor

Hi {{FIRST_NAME|readers}},

Quick ask before we get into it this week.

Google's AI now answers procurement questions for people, saving them the step of clicking onto webpages 😬

Google builds those answers from whatever webpage ranks, including the fake vendor rankings, the recycled LinkedIn takes, and the "thought leadership" written by whichever model was cheapest to run that morning.

Everything we publish is the opposite of that.

Practitioner-written. Buyer-protective. Actual opinions from actual humans, formed on actual projects with actual procurement leaders.

If that's why you're here, please take 15 seconds to mark Pure Procurement as a Trusted Source in Google.

One checkbox, and it tells Search, Discover, and Google's AI answers to put our work ahead of the slop.

Onwards!

📰 In this week’s edition:

Note: Some of the content listed above is only available in the email version of this newsletter. Don’t miss out! Sign up for free to get the next edition.

(And Why Every S2P Vendor Just Shipped One)

In June, the last of the Top 8 S2P suites in the Gartner Magic Quadrant quietly shipped its "AI Agent Studio." Every serious procurement platform on the market now has one.

So it's a good week to ask some questions...👀

Let's get into it.

An AI Agent Studio is a no-code or low-code environment, embedded inside an enterprise platform, where your team builds, configures, and orchestrates LLM-supported business processes across embedded and connected stakeholders, systems, and data.

Instead of hard-coding a fixed process (the workflow builders of yesteryear), you drop LLM-powered steps into a workflow and chain them together toward an outcome you’re looking to achieve.

The "studio" part is what matters. It's not just embedded AI functionality. It's a place where your practitioners design and wire up these processes, and where governance controls decide where the model gets to act on its own, where a plain old non-LLM algorithm does the job, and where a human has to stay in the loop.

That last part is the whole ballgame. More on that in a second.

If the term feels slippery, that's because the category hasn't settled on a name yet.

What one vendor calls an "Agent Studio," the next calls an "Agent Builder," an "Orchestration Studio," or an "Agentic Platform." Same capability, different sticker:

The labels differ. The underlying capability shouldn't. So, the skill worth building is recognizing the thing regardless of what a vendor stamps on the box.

If you take nothing else from tonight’s note, take this.

Nearly every vendor now ships LLM features baked straight into their workflows. Those features are useful. But they are the vendor's processes, running the vendor's logic. Not yours.

A studio is different. A studio hands you the keys to the building.

→ Embedded AI = you consume someone else's automation

→ A studio = you author your own

That shift, from renting a fixed LLM-supported process to authoring your own, is the real story behind all the naming noise. It's also the exact capability to pressure-test in a demo before you take any "Studio" label at face value.

Ask the vendor to build a net-new agent, live, in front of you. Not run a pre-canned one. Build one. What you're watching for is whether your team could do that after go-live, or whether you'll be filing a ticket every time you want to change a step.

And it's not just the Magic Quadrant crowd

Another important point is that plenty of vendors outside the Gartner MQ for Source-to-Pay Suites shipped this capability too, and in a lot of cases they did it months (sometimes years) ahead of the suites.

ConvergentIS, LevelPath, Pivot, ORO Labs, Vertice… They all have the equivalent of an AI Agent Studio… And I’m sure I’m forgetting lots of others, let alone the capabilities delivered by best-of-breed vendors.

The point stands regardless of the final list: the "Studio" capability is not a Gartner MQ suite-only club, and the newer players often got there first.

Because practitioners are figuring out the thing the early AI hype conveniently skipped:

The more you ask a single LLM call to do, the higher the odds it hallucinates. So, getting reliable results at scale means being deliberate about how, when, and why an LLM enters your process at all, and combining it with traditional, deterministic automation along a pre-defined workflow.

Which, when you say it out loud, sounds familiar.

Workflow builders have existed for decades. From where I sit, an AI Agent Studio is just the next iteration. Same instinct (design the process, then automate it), new tool in the box (an LLM that can handle the fuzzy steps a rules engine never could).

Two things.

None of these is automatically right or wrong. But every one of them affects your future flexibility, and most of them never come up unless you ask.

If you're serious about building an agentic procurement organization, an AI Agent Studio isn't optional. You need one…

But landing a decent, flexible studio is only the starting line.

The real work (the part that actually separates the teams getting value from the ones stuck in “proof of concept”, automating their own confusion) is developing agentic business process management as a capability inside your team. That's how you get consistent results at scale. A good tool may help but it doesn't do that for you. Your people do.

The studio just gives them somewhere to do it.