Acumatica made 2026 R2, the latest release of its cloud ERP platform for mid-market companies, generally available on October 1. The press release groups the update into four areas: embedded AI, analytics, user experience, and industry editions.
Much of the AI work addresses two questions ERP teams face once AI moves past pilots: what the system may do on its own, and what data outside AI tools can see. Acumatica routes both through controls already in the ERP, including role-based access rights, published inquiries, and activity logs.
Jon Pollock, chief product officer at Acumatica, said growing businesses want “practical AI solutions” instead of more AI experiments.
How AI Agents Work in 2026 R2
According to Acumatica’s release overview, users create agents with plain-language instructions and save them for reuse. An agent can start automatically from a Business Event, an Import/Export Scenario, or an existing workflow. It draws context from Generic Inquiries, Acumatica’s configurable data queries, takes a defined action on eligible records, and can hand the task to another agent or automated step. Acumatica’s examples include adjusting prices to meet target margins, qualifying leads, and assessing credit risk.
AI Assistant also takes on harder financial questions. It can analyze document history to explain how a balance changed across prior periods or on specific dates. Answers can be saved as dashboard widgets, and users can review the source inquiry behind each one.
Controls sit around both features. Agents run under existing access rights. Administrators can mask selected fields before AI Automation sends a record to a large language model, with original values restored inside Acumatica. Processing logs show provider-reported token counts for each automation, and the release adds warnings when AI-unit consumption reaches set thresholds.
How Outside AI Tools Reach ERP Data
The release also changes where analysis runs. The press release says reporting and analytics now use a dedicated, performance-optimized copy of ERP data, so reports do not compete with order entry and invoicing for system resources. Acumatica describes this native data warehouse as a foundation it plans to extend in future updates. Generic Inquiry Optimization stores the results of large or frequently used inquiries, refreshed on a schedule or on demand, for reuse across dashboards, OData feeds, pivot views, and MCP connections.
The new Model Context Protocol (MCP) servers let administrators expose published Generic Inquiries as callable tools. Assistants such as Claude, ChatGPT, and Gemini then get read-only access to selected Acumatica data without exports. Each request runs under the user’s access rights and is logged. A separate DAC-level security setting limits which underlying data objects business users can query when building their own inquiries.
Industry editions get their own additions. Construction gains parent-child projects, S-curve cost forecasting, and a vendor portal that the community announcement calls the first step toward a broader Construction Exchange. Manufacturing adds an Advanced Planning and Scheduling engine, and distribution adds Available-to-Promise dates on order lines. Acumatica says payroll calculations and batch releases run up to 63% faster.
The Modern UI is now the default. Classic-only screens remain available with advance warnings until the 2027 R1 release retires them.
What This Means for ERP Insiders
Treat the inquiry layer as the AI boundary. In 2026 R2, agents, AI Assistant, and MCP connections all draw on Generic Inquiries. Reviewing which inquiries are published, and to which roles, gives ERP teams one place to decide what AI can see.
Define agent triggers before agent actions. Agents can start from business events and pass work along a chain. Mapping which events start which agents, and who reviews the actions they take, keeps multi-step automation traceable when an outcome is questioned.
Build the Classic UI migration into 2027 planning. Classic-only forms, dashboards, pivot tables, and customizations are scheduled for retirement in 2027 R1. Inventorying them during the 2026 R2 cycle gives teams time to test replacements before the cutoff.
