It's easy to build an AI agent and even scale them. It's much harder to orchestrate them to drive returns. AI agents need a manager pronto.

Those are the takeaways from Boomi CEO Steve Lucas and IBM's Bruno Aziza, Group VP of Data, BI and AI. Aziza also is a member of IBM Ventures' investment committee.

Both tech leaders spoke at Constellation Research's Connected Enterprise conference where AI agent sprawl and orchestration was a recurring theme.

Boomi calls BS on inflated AI agent numbers

Boomi CEO Steve Lucas said enterprises are moving to AI agents as they shift from deterministic processes to probabilistic processes, but be wary of inflated claims about how many are being deployed.

"We live in this world called reality," said Lucas. "We've deployed roughly 50,000 agents in our customers that are real with code and grounding for business processes. That's not millions and anyone that tells you millions is full of shit. It's not true."

Lucas added that AI agents need to be judged on returns and real business process impact. "We're seeing real ROI," said Lucas, who noted Boomi is deploying internally and with customers.

He added that there's a ground game to consider. "When we walk into a client today we start with a workshop. You can build a new agent with a chat interface, but do they have the right data and process?" asked Lucas.

Next up will be the orchestration of agents, which is often more of a vendor talking point. "How do I monitor for effectiveness and ROI? You have to determine the ROI and whether agents are overlapping all day," said Lucas.

Boomi has an agent builder that includes governance, but will extend the tool to monitor ROI and orchestrate multiple AI agents. "What we're going to be announcing next year is the multi agent orchestration extension. We're not building an agent control towers. We don't just watch AI agents work. You need to orchestrate multiple agents and figure out do you have the right model and what are the costs."

Lucas said 2026 will be about creating an AI activation layer and moving beyond automation. "It's not just about creating agents it's going to be about orchestrating," said Lucas.

IBM's Aziza: AI agents have no manager right now

Aziza said it's easy to create an agent--he has five or six GPTs to serve up answers to his kids.

He said:

"Building an agent of agents is really simple. The problem stuff that's going to get us in trouble is you're going to have multiple agents built across multiple platforms, and no single vendor will have the incentive to just help you manage and  orchestrate across multiple platforms. AI has no manager right now."

The state of play today is the following:

  • Vendors are making it easy to build agents for developers as well as business users.
  • Enterprises can choose to stay on one platform for agentic AI, but that's limiting.
  • The challenge will be finding the vendor that's horizontal and "coming to help you orchestrate and operationalize these agents."

IBM TechXChange 2025: Big blue connects agentic AI, mainframe dots, partners with Anthropic

"The key word for us in the next 10 years is going to be multi-agent orchestration," said Aziza.