Splash 2026: SensibleAI Agents Go GA and Finance Gets an Agentic Layer

At its Splash 2026 user conference in Orlando this week, OneStream announced general availability of SensibleAI Agents and unveiled a new Finance Agentic Layer — the two moves that turn last year's AI story from demos into shipping product for the Office of the CFO.

Four Agents, Shipping Today

The GA release covers four native agents — Search, Finance Analyst, Deep Analysis, and a new Forecast Agent — that run directly on the OneStream platform and integrate into the Microsoft 365 suite for reporting, financial analysis, and informational search. The Forecast Agent is the most interesting of the set for planning teams: ask questions about your forecast in plain language and get answers in financial terms, grounded in live forecast data rather than a stale extract. These are positioned as task-enabled workers, not chatbots — shipping, not previewed.

The Agentic Layer Is the Bigger Story

The new Finance Agentic Layer opens OneStream's capabilities to third-party AI tools, so finance teams can forecast, guide, and report on the business from Copilot, Claude, ChatGPT, or Gemini. The governance design is what makes this credible in a finance context: every interaction through the layer is authenticated against the user's OneStream identity and enforced against their existing role-based permissions, with administrator visibility into usage and activity. In other words, the agent your analysts use inherits exactly the security model your consolidation data already has — no side-channel exports, no shadow copies of the trial balance in a chatbot's context window.

Our Take

This is the announcement that makes AI governance a OneStream configuration question instead of a policy memo. If your organization has been holding AI at arm's length because finance data couldn't leave the platform, the calculus just changed: the platform is now the control point. The practical to-do list follows from that — get your role-based security model genuinely clean (agents will faithfully expose whatever access users already have), get on a platform version that supports the agent framework, and pick one workflow, like forecast Q&A or close status reporting, to pilot before rolling agents out broadly. Teams that treat this as a security-design exercise first and a productivity tool second will get the value without the surprises.