Dive Brief:

• Finance software maker OneStream on Tuesday announced new agentic artificial intelligence capabilities, including a feature that allows interoperability with third-party systems such as OpenAI’s ChatGPT, Anthropic’s Claude and Microsoft’s Copilot.
• At the center of the rollout is what OneStream calls its “finance agentic layer,” which is designed to let its AI agents work with external tools.
• “It's no shock that we're starting to see more and more customers make a choice of, ‘I’m all in on ChatGPT or Claude or Microsoft Copilot,’” Andrew Shea, a data scientist and software engineer at OneStream, said in an interview. “That's something that we've rapidly responded to.”

Dive Insight:

Enterprise software providers are accelerating efforts to embed agentic AI across corporate finance and other business functions, raising broader questions about how governance, auditability and interoperability will function at scale.

Last week, SAP launched a new AI initiative dubbed “Autonomous Enterprise.” As part of the rollout, the company said it was enabling agent-to-agent interoperability with external applications.

A report by Constellation Research characterized SAP’s move as “table stakes,” arguing that “all agents will have to use common protocols, share data and be interoperable.”

OneStream’s latest announcement comes about a year after it launched a series of AI agents, including a “finance analyst agent” and “deep analysis agent.” On Tuesday, the company added a new “forecast agent” designed to allow users to ask questions about forecasts in plain language.

Previously, OneStream focused on deploying only “native” AI agents operating within its own platform. The company’s new approach is designed to let those capabilities work with external systems while preserving controls required by finance teams.

In a Tuesday press release, the company said its new finance agentic layer uses a so-called open model context protocol to give any AI tool “secure, governed access to OneStream data with the financial logic, auditability, and permissions and audit trails that finance requires.”  

“The idea is to make sure OneStream’s financial intelligence layer can surface wherever the customer is working,” Shea said.

He described MCP as a “USB plug for AI systems to talk to one another,” enabling organizations to connect OneStream’s agentic tools with any external AI ecosystem.

“When ChatGPT is asked for a dashboard showing trends, variances, and forecast projections, it retrieves reports from OneStream using the Finance Agentic Layer and produces a C-level output in minutes,” the company said in its release.

Every interaction through OneStream’s finance agentic layer is authenticated against the user’s OneStream identity and “enforced against their existing role-based permissions,” ensuring that users working in external AI tools only access data they are authorized to see, the release said.

The company added that administrators retain visibility into system usage and activity within OneStream, “providing the control and auditability required for financial operations.”

“If the numbers aren’t grounded in a trusted source of truth, you are going to end up getting the wrong numbers, and trust is going to break down very quickly,” Shea said.