OPERATIONS · 5 MIN READ
Agents on top of your ERP, without a migration
You do not need a new system to get AI into operations. Agents can read from and write to the ERPs you already run.
LAUNCH AGENTS ·
The migration myth
Many operations leaders assume AI must wait for a new ERP. The current system is old, customised or one of several across business units, and the project to replace it is always eighteen months away. Meanwhile, the people running operations keep copying data between screens and spreadsheets.
AI agents do not need a clean, modern system of record. They need access to the data and a safe way to act on it.
How agents connect
Agents connect to existing systems through connectors. The Model Context Protocol (MCP) is an open standard that lets an AI model use tools and data sources in a controlled way. A connector exposes specific capabilities, such as reading stock levels, looking up an order or creating a draft purchase order, with permissions set per action.
Where a system has an API, the connector uses it. Where it does not, agents can work with exports, reports and spreadsheets that teams already produce. Several ERPs across business units can sit behind one agent, which is often the first time leadership sees a single view of operations.
Read, draft, approve, write back
- Read. The agent pulls the data it needs: orders, stock, supplier terms, customer history.
- Draft. It prepares the action: a replenishment order, a transfer between warehouses, a corrected entry.
- Approve. A person reviews and approves, individually or in batches, depending on risk.
- Write back. The approved action is written to the ERP, with an audit trail of what was done and why.
Where this pays off first
Good starting points are workflows where people move data between systems or make routine decisions from data that already exists: order entry from emails and spreadsheets, inventory replenishment, supplier quote comparison, and order status enquiries. Each has a clear metric, such as order cycle time, fill rate or response time.
What it does not replace
Agents do not remove the need for good master data or sound controls. They often expose data problems faster, which is useful. And when an ERP replacement does happen, the agents and the redesigned workflows carry over, because they were built around the work, not the system.
Key takeaways
- AI in operations does not have to wait for an ERP migration.
- Connectors such as MCP let agents read and write existing systems with per-action permissions.
- The pattern is read, draft, approve, write back, with a full audit trail.
- Start where people move data between systems or make routine data-driven decisions.
Frequently asked questions
Can AI agents work with legacy ERP systems?
Yes. Agents connect through APIs where available, or through exports, reports and spreadsheets where not, and write approved changes back with an audit trail.
What is the Model Context Protocol (MCP)?
MCP is an open standard that lets AI models use external tools and data sources in a controlled way, with permissions defined for each capability.
Do we need to clean our data before using AI agents?
Reasonable data quality helps, but you do not need a perfect dataset to start. Agents often surface data issues quickly, which helps prioritise fixes.