Last week, the Nibble team attended the Konnect House “Agentic AI in Procurement” event in New York, where procurement leaders and technologists gathered to discuss the future of AI-driven procurement. The room was full of big ideas… Autonomous agents, AI negotiation, procurement orchestration, and the promise of doing more with fewer resources.
But beneath the buzzwords, three themes kept surfacing again and again. If you weren’t there, here are the messages that mattered most.
1. The hype around agentic AI is far ahead of real adoption
Early in the event, panel host Ana Paula Zamorano asked: “How many of you are exploring agentic AI?” – Lots of hands went up, of course. But then came the follow up:
“How many of you are actually using agentic AI in a project today?”
Almost every hand in the room dropped – I think I counted three hands left up! That moment captured the current reality of AI in procurement: interest is huge, but real deployment – particularly on anything that could interact with suppliers externally – is still rare.
Several speakers highlighted why.
Michael Shields of Tropic reminded the audience that many so-called “agents” in the market aren’t really agents at all. True agentic systems should be able to handle multi-step tasks, make decisions, and retain memory across workflows. In other words, they should produce meaningful outcomes even when a human isn’t guiding every step; but most organizations aren’t there yet.
In industries like healthcare or pharmaceuticals, the challenge is even greater. Regulatory pressure and risk mean teams often use AI as decision support rather than full automation, with humans still validating outputs before anything goes live.
There’s also a wider industry reality to contend with: according to Gartner, around 40% of AI projects being implemented today will fail by the end of next year. That statistic came up in conversations about stakeholder buy-in. When so many initiatives stall or die, convincing leadership to commit to AI programs becomes harder. Teams need to feel empowered to be able to demonstrate ROI quickly and easily when it comes to agentic AI.
Agentic AI is attracting enormous attention, that is certain, but it seems we’re still in the early innings of real enterprise deployment. There’s a lot of exploration going on, but less action, leaving space for truly innovative enterprises to capitalize early.
2. Procurement teams are experimenting everywhere, but expectations need resetting
Another recurring theme was what Arie Barendrecht, from Omnea, described as “the AI expectation gap” when talking about automation. Executives want transformation quickly. Stakeholders are increasingly asking procurement teams to “bring AI in NOW.”
…But the reality of implementing AI inside large organizations is much slower. Many companies are still experimenting with small proofs of concept rather than enterprise-wide deployments. And several speakers argued that this is the right approach.
Instead of jumping straight to fully autonomous systems, organizations should start with small, targeted use cases to prove value that can be implemented quickly with low integration, using simple workflows where AI can demonstrate clear value.
Examples mentioned throughout the event included:
- Automating tail spend negotiation
- Using AI to rank RFQs and flag exceptions
- Supporting ROI calculations
- Streamlining routine procurement decisions
Many procurement leaders spoke about wanting agentic AI to free their team to focus on strategic decisions and supplier relationships. That reflects a broader mindset shift.
For many leaders, the real ROI of AI isn’t simplt the cost saving; it’s freeing up time for higher-value work.
There is, admittedly, a challenge here: if AI handles more analytical tasks, how will younger employees learn those skills in the first place? That is perhaps a bigger question for a separate article… Automation may create more efficiency, but it can also remove opportunities for learning.
3. The “build vs buy” AI debate is heating up
Perhaps the most practical conversation at the event centered on a familiar question in enterprise tech (and one we held a webinar on recently)
Should procurement teams build AI tools themselves or buy them from vendors?
The rise of agentic AI has made this question more relevant than ever. Some procurement teams are now experimenting with building their own agents internally, especially for early prototypes. With modern AI tools, creating a simple proof of concept can be surprisingly quick and inexpensive. So, if a procurement team approaches their CFO with a new idea, one of the first questions they may be asked is, “have we tried to build this ourselves?”
In heavily regulated industries like healthcare, this approach can also be necessary. Organizations may need to build internal solutions simply to meet compliance requirements.
But several speakers cautioned that building AI isn’t free in the long run. Aubrey Zimmerman, from Ramp, in particular raised the point: once you create something internally, you also have to maintain it, scale it, secure it, and integrate it into real enterprise workflows. Do you have the bandwidth for this?
That’s why some companies still lean toward buying specialized software.
Victor Scarante, Head of Procurement from Hershey, described their approach as buying rather than building wherever possible, arguing that the time saved can far outweigh the cost of the software.
Others suggested a hybrid strategy:
- Build for simple, repeatable internal tasks
- Buy for complex or specialized capabilities
Either way, a key piece of advice emerged: avoid locking into long-term commitments too early. Start with small pilots, test solutions with messy real-world data, and expand only once value is proven.
Final takeaway: the AI conversation is still catching up with reality
If there was one feeling that defined the event, it was this: AI in procurement is still in its experimental phase.
There’s enormous excitement about agentic systems, autonomous negotiation, and AI-driven procurement workflows. But many organizations are still figuring out what actually works.
The companies making the most progress aren’t chasing the biggest promises. They’re testing small use cases, sharing early successes, and building momentum one project at a time.
Agentic procurement will eventually transform how teams operate, that much seems inevitable. But for now, the real work is much simpler, and much harder: turning experimentation into real, repeatable value.
