"I believe we are in the midst of a revolution — and I think in a revolution, we as a vendor also want to invest a lot in educating the community on what's really changing and how it's changing."
In this first episode of The Agentic Procurement Diaries podcast, I sit down with Aatish Dedhia, Founder and CEO of Zycus. We discuss why the series exists and what they both hope it does for the procurement community.
Aatish leads product at Zycus, and he has spent more than two decades building procurement technology from the inside. He has watched the category move from paper to digitization to, now, agentic AI, and he talks about each shift with the specificity of someone who lived through the failures as well as the wins. In this episode, he walks through what changed, what broke, and what he would tell a procurement leader wondering whether any of this is real.
THE INHERITANCE
Aatish did not go looking for procurement. In the early 2000s his team had a machine learning technology for aggregating and classifying internet data, and a VC asked if they could do the same for procurement catalogs. They ran a POC, and the journey began.
He arrived just as procurement's digitization started. That first wave standardized processes and centralized data, but as Aatish points out, it did not automate. It made people follow the same steps and made data visible. Contracts came out of drawers. That was the status quo he built on.
THE BET
Two years ago, Zycus bet on building a lot of point agents sprinkled across its source-to-pay product. They shipped them, and customers did not see the ROI.
Aatish pivoted quickly toward true agentic flows, systems that start with the end outcome and work backwards, lining up and coordinating the agents needed. Without that pivot, he says, they would not be at version 5 today. He frames it as the right kind of failure: failing while exploring a new area, not failing at something simple.
THE AGENTIC QUESTION
Aatish is willing to let an agent run with no human in the loop, but only on the right work. For a tactical buy he has seen AI handle again and again, he will say go ahead and do it autonomously.
For a first-time services RFP, say cybersecurity, he wants a human involved in the RFP and the final evaluation. His point on agentic sourcing is that you decide where the human enters, and everything sits on a deterministic system of record that gates the agents and blocks anything really wrong from reaching the ERP.
THE RECEIPT
What worked: inside Zycus, AI drove a 45 to 50 percent efficiency gain over two years, with half the team doing significantly more work across products, sales, and code.
What did not: point agents alone created agent sprawl and missed the big ROI.
And AI put on a broken process fails. A customer applied agentic intake to a rule requiring approval for everything above $500, and AI simply did what the broken process told it to. One limit remains honest: audit trails show what an agent did, but explainability, the why behind a recommendation, is still hard.
.png)
