From conversational helpers to finance AI agents
Finance AI agents are software systems that combine large language models with enterprise context to execute ERP-adjacent accounting work, such as reconciliations, close tasks, and intercompany postings, under explicit human oversight and within defined control frameworks. This marks a shift away from AI chatbots that only provide explanations or summaries. At events like Sage Future in San Francisco, the discussion has moved from pilots to production: finance teams want tools that remove repetitive work from their day, not more conversational interfaces layered on top of the ERP. Nominal’s approach reflects this change by placing agentic performance management alongside the ERP, where agents follow standard operating procedures instead of inventing new ones. For high-volume, multi-entity businesses, this kind of autonomous accounting workflow support promises time savings, provided that execution remains traceable, governed, and aligned with existing finance controls.
Why explanation is not enough in accounting workflows
Chatbots improved access to finance data, but they left a gap: they could explain variances or summarize ledgers without closing any books. In accounting, value comes from completed steps—journal entries posted, reconciliations cleared, intercompany transactions matched—so finance AI agents now focus on execution rather than conversation alone. Nominal positions its platform as an execution layer that mirrors how finance teams already work, following standard operating procedures and ERP configurations instead of bypassing them. This agentic AI execution model still keeps humans in charge: teams define guardrails, approve higher-risk actions, and retain audit trails for every task. The goal is not a fully autonomous accounting workflow that cuts people out, but a division of labor where AI handles repeatable tasks at scale while controllers and accountants retain judgment, sign-off, and accountability for final financial outcomes.
The execution gap: from agent intent to ERP automation
As enterprise platforms add agentic capabilities, the main question has shifted from intelligence to execution. SAP’s announcements around Joule Work, more than 200 agents, Company Memory, and an Anthropic partnership show how quickly agents are being woven into business applications. According to Charles Crouchman, Chief Product Officer at Redwood Software, “the question has moved to, ‘Can AI actually execute inside my business?’” His concern is the execution gap that appears when probabilistic AI actions flow into long, tightly coupled processes like financial close, MRP runs, or billing cycles. A single plausible but inconsistent action may not fail loudly; it can propagate downstream, turning promised efficiency into remediation work. Redwood’s response is to treat existing workload automation as a governed execution fabric, so agent intent is translated into deterministic ERP automation with timing, dependencies, and approvals already encoded before systems of record are touched.
Automation with accountability: governed agentic AI execution
Both Nominal and Redwood focus on the same tension: how to expand ERP automation without losing control over how financial data changes. Finance AI agents capable of initiating journal entries or triggering ERP automation must operate within clear governance structures. In SAP environments, AI Agent Hub and Company Memory form an intelligence layer that coordinates agents and preserves process knowledge. Redwood’s layer begins at the handoff point, where agents call RunMyJobs workflows as governed tools with hard guardrails. This design keeps the autonomous accounting workflow concept grounded in accountability: agents can propose and trigger work, but predefined automation paths, approvals, and logs ensure traceability. In lower-risk processes, native ERP automation may be enough; in complex hybrid estates, organizations increasingly see the need for a separate execution layer that constrains agentic AI execution before financial records are updated.
Context and integration as the next finance AI advantage
The emerging pattern is clear: raw AI model strength is no longer the main differentiator for finance teams. Context, integration, and reuse of existing ERP automation matter more. Many organizations have years of logic encoded in workflows, job schedules, and integrations that govern how financial close, intercompany processing, and data movements run across SAP S/4HANA, older ERP systems, and cloud applications. Crouchman describes this as solving the cold-start problem by turning automation libraries into tools that agents can call “from their first interaction.” Nominal follows a similar philosophy, remaining ERP-agnostic while embedding its agents into finance operations and standard operating procedures. The competitive edge now lies in how well finance AI agents plug into this landscape: understanding company memory, using existing ERP automation, and keeping humans in the loop, so that automation scales while governance and auditability remain intact.
