The race is on to parent your soon-to-be sprawling set of AI agents with orchestration layers, AI studios and a bevy of tools to build, deploy, manage and optimize this digital workforce. Being in the pole position as an agentic AI platform is going to be lucrative for a few enterprise vendors, but choose wisely.

A set of announcements in recent days featured how the race to be the platform of AI agents is ramping.

ServiceNow outlined plans for CRM, a data partner network, an AI control tower, AI agent management tools and other key features to extend the company's workflow and automation platform to AI agents. ServiceNow CEO Bill McDermott said: "We're going to bring AI agents to every corner of your business."

A screenshot of a computer AI-generated content may be incorrect.

IBM at its Think 2025 conference outlined its AI agent orchestration platform. IBM is looking to build, deploy and manage AI agents for multiple use cases that create a "system of intelligence."

Ritika Gunnar, GM of Data & AI Software at IBM, said: "We built systems of records. These things are like our digital vaults. We connected them through systems of engagement. We gained understanding of them with our systems of insight and our analytical dashboards, which were grounded with our warehouses and our data lakes. But knowing isn't enough. The next big step is systems of intelligence powered by AI agents. These aren't just dashboards. They're doers. They act autonomously orchestrating workflows across your enterprise."

Gunnar noted that "the explosion of AI agents across the enterprise holds immense promise, but let's be real for a minute, it also creates significant complexity."

A screenshot of a computer AI-generated content may be incorrect.

UiPath launched its automation platform, which includes AI agents as part of a holistic approach that includes RPA, applications and the enterprise estate.

These are just a few examples of how enterprise vendors are looking to manage AI agents--their own as well as third parties. Boomi, which has its Boomi World conference this week, is also a key player in the AI agent orchestration race. Obviously, Salesforce with its Agentforce effort has another platform play for AI agents as does SAP. And of course, all the hyperscale cloud providers--Amazon Web Services, Microsoft Azure and Google Cloud--have platforms and the ambition to manage agents across enterprises.

With that backdrop, it's worth thinking through what you want in an AI agent orchestration layer. Here are some of the requirements that are bubbling up from CxOs.

Horizontal approach

Enterprise AI buyers don't want a portfolio of platforms managing their AI agents. Yes, you're likely to have agents defined by software categories such as CRM, HCM and ERP, but the aim is for enterprises to have an orchestration to manage them.

It's no secret that system integrators have done their best so far at building AI agents. These integrators are accustomed to working across systems, data silos and vendors. The base requirement for any agentic AI platform is the ability to work across systems, data stores and functions.

PepsiCo has more than 1,500 bots, agents and assistants across the company. Magesh Bagavathi, SVP, Chief Data and AI Officer a PepsiCo, said at IBM Think that the company has "a platform centric approach."

Bagavathi looks at two kinds of platforms. First, PepsiCo has built an agent orchestration platform on IBM's watsonx platform. "We're really looking at platforms for our business end to end," he said. PepsiCo started with proof of concepts and then moved to production by processes including accounts receivable and accounts payable with the aim of moving throughout the company's value chain.

Get the Constellation Insights newsletter

PepsiCo is also building its own internal abstraction layers for agents, AI and data with IBM Consulting.

Cindy Hoots, Chief Digital Officer and CIO at AstraZeneca, said during the ServiceNow Knowledge 2025 keynote that the company used ServiceNow to create a unified platform to save time and money across everything from HR to R&D. Like PepsiCo, AstraZeneca looked to automate workflows and processes with AI.

"We're finding agents are really embedded in every part of our company and using them to transform patient care as well as drive healthcare innovation," said Hoots. "It's just something that we've kind of built into every aspect of the company for research and development all the way down, and it's helping us have much more autonomous ways of working. They're helping the team to really focus on those higher value activities."

IBM and ServiceNow historically work across systems, but there are a bevy of others. Horizontal characteristics can be found in hyperscale cloud vendors (think Amazon Q), integrators and platforms that historically connect systems (Boomi, UiPath, process mining vendors and other firms).

Enterprises may choose to build their own platforms too.

Neutrality

Neutrality is a tricky topic because a vendor that has historically worked across systems naturally wants you to consume as much of its platform as possible.

Vendors that are coming from a SaaS orientation and working to become more of a broad platform have a lot to prove. Salesforce's Agentforce makes the company's various clouds more of a platform, but perceptions will take time to change. SAP can launch Datasphere and partner with Databricks to connect third party data sources, but there's little doubt that it wants enterprises on its cloud and data stores (all connected by its Joule agent).

The big vendors can be seen as neutral players that ultimately drive costs down, but CxOs will remain skeptical.

Now enterprises can opt for smaller vendors that are good at integrating platforms--think Boomi or UiPath--and get neutrality. But these smaller players could always simply be acquired by larger vendors. MuleSoft, an API platform that was acquired by Salesforce, is an example. A neutral vendor today may not be neutral tomorrow.

ServiceNow's Amit Zavery, Chief Product and Chief Operating Officer, drove home the neutral platform argument. ServiceNow provides model, infrastructure and data neutrality. IBM's CEO Arvind Krishna had a similar riff.

Connectors, data integration

It's not hard to find CxOs who note that AI agents are largely glorified souped up APIs. As a result, you'll want vendors that have an API heritage as your AI agent control plane.

For instance, Boomi has evolved from being a pure play iPaaS vendor to a broader platform covering AI, AI agents, data app integration and automation.

Connectors are the base layer for any AI agent orchestration effort. If a vendor isn't supporting Model Context Protocol (MCP) or Google Cloud's Agent2Agent standards they aren't likely to be a good choice. Microsoft recently became the latest to support Agent2Agent and noted that AI agent interoperability is critical.

This same approach applies to data integration. Yes, it would be swell if all your data were in one lakehouse and organized. It is also fantasy. As a result, the ability to connect and access data wherever it resides is going to be critical to AI agents.

Given the importance of data integration it's not shocking that ServiceNow launched its Workflow Data Network, which is powered by RaptorDB Pro. These connections include Snowflake, Databricks, Boomi, SAP, the hyperscale cloud providers, Jira and Workday.

A screenshot of a computer AI-generated content may be incorrect.

Process and use case knowhow

What has been most shocking about the AI agent hype is how little play process gets. If the process isn't continually optimized, AI agents are likely to simply take the screwed up processes enterprises have today and scale them.

Microsoft, ServiceNow, SAP and UiPath talk process and use cases more than the rest of the field. IBM talked about 150 pre-built agents in watsonx Orchestrate across domains ranging from HR, sales, procurement and IT and a catalog of agents for partners.

ServiceNow is looking to a suite of AI agents as an overlay of systems in HR, procurement and finance.

A screenshot of a green and white screen AI-generated content may be incorrect.

Bottom line: Your AI agent layer should have some serious process intelligence behind it.

Integration skills

Today, AI agents don't exactly work well out of the box. There are a few prepackaged AI agents within SaaS silos, but integration across systems and data silos matters.

This enterprise-specific integration is one reason why it's quite possible that this agentic AI layer is built by companies with the help of consultants such as Accenture, IBM, Infosys, Cognizant and a bevy of others.

Start with the business problem you're trying to solve, think through the integration and then platform options.

Business Research Themes

Data to Decisions, Future of Work, Innovation and Product-led Growth, Next-Generation Customer Experience

Roles

Chief Information Officer