AI can live up to the hype. The right data makes all the difference.

There's a growing gap between what agentic AI promises and what it delivers today. I think the companies still standing in 20 years will be the ones that close that gap, not the ones that profit from inflated expectations until reality catches up with their customers.

That's what interests me about Gartner's Hype Cycle for Agentic AI. Agentic AI is sitting at the peak of inflated expectations, but the curve tracks technologies and expectations, not the companies selling them. And the two won't necessarily follow the same trajectory.

There's a limit to how long vendors can sell what AI could do before customers start asking what it actually delivers. An impressive demo might win a deal. An exciting product announcement might justify a premium. But when the promised results don't materialize, customers leave.

The technology can survive the hype. But that doesn't mean every business built around it will.

I don't care how hyped AI is. I care about what it delivers.

There's nothing wrong with using an 'overhyped' technology. If an agent saves a procurement team significant time and money, does its position on the Hype Cycle really matter? Probably not, IMHO.

And there's evidence that the biggest returns will come from agents built to solve specific problems. After examining 100+ agentic AI deployments, Gartner predicts that 80% of tangible agentic AI ROI will come from specialized, domain-specific agents by 2028.

We built an agent for a customer that benchmarks every purchase request against the company's internal data. It helped generate almost 10% savings. That result is far more interesting to me than the number of agents we've launched.

Of course, giving an LLM a procurement-specific name doesn't make it 'custom'. It needs to understand the company's procurement context, access clean and complete data, and produce trustworthy outputs.

The winners in enterprise AI won't be the ones with the most agents. They'll be the ones whose customers have the most to show for them.

Nobody needs another AI agent that only works in a demo.

There's a lot of exciting innovation happening in ProcureTech right now. But AI has also created an incentive problem. Announcing ten new agents might make the next fundraising deck more exciting. It tells us very little about what those agents can achieve for customers.

I believe the best AI agents will come from working closely with customers, not guessing what they need. That's why we built Pivot's AI Studio. Our forward-deployed engineers work with procurement teams to understand their problems, identify where AI can help, and build customized agents. Then we test them and measure the results.

Some customers come to us with a specific use case. Others know something is inefficient but have no idea how AI could improve it. In both cases, it's our responsibility to bring the procurement process expertise and engineering capabilities needed to solve the problem.

Customers shouldn't have to figure out how to make our AI valuable to their business. That's our job. And I think the companies that take that responsibility seriously will have a lasting advantage.

The gap between expectations and reality starts long before the agent.

But even the best engineers can't build reliable agents on data that doesn't exist. And that's where I think a lot of the industry's AI ambitions run into a much less glamorous problem.

Vendors are promising autonomous procurement while many companies still struggle to capture basic purchasing data. The problem often starts with inadequate procurement software. Employees avoid complicated interfaces, purchases happen outside established workflows, and important data never reaches the system. Then we connect an AI agent and expect miracles.

We've built Pivot around three layers: a system of record, a system of engagement, and a system of agentic action. The first gives us structured procurement data. The second makes it easy for employees to contribute to that record. The third lets agents act on it.

We've invested heavily in all three because even the most impressive model cannot compensate for a contract that was never captured or an outdated supplier record.

Better models will help. But better foundations make their capabilities applicable to actual business problems. That's how we intend to narrow the gap Gartner describes, one measurable outcome at a time.

An AI wrapper is a feature. Customer trust is a business.

Connecting an LLM to a procurement platform has never been easier. And as models improve, the AI features that look impressive today will become easier for competitors to replicate.

That raises a question I think every enterprise founder should be asking: what will customers still be paying you for when the technology you're showcasing today is available to everyone?

My answer comes back to the things we've been building at Pivot. Reliable procurement data. Infrastructure that lets agents act on it. And people who understand customers' problems well enough to build the right solutions.

Those capabilities take time to develop. They also reinforce each other. Better data makes more ambitious AI applications possible. Working closely with customers helps us identify which applications are worth building. And when those applications work, customers have a reason to trust us with more of their procurement operations.

We're building Pivot for the next 20 years. I wouldn't bet that future on having the most agents. I'd bet it on knowing how to make them work for our customers.