Hesham Elmahdy

What does the agent need to know?

Can an AI agent safely read my company's ERP system?

The short answer

Yes, if you treat the agent like a new employee you do not fully trust yet. Give it its own account, let it read only, and let the ERP system itself decide what it may see. The danger is not that the agent reads your data. The danger is giving it more access than any person in that role would get.

Why connect it at all

An agent that cannot see your ERP system answers from guesses. Your clients, prices, stock and invoices all live there. So for most companies the real question is not whether to connect the agent, but how.

The agent gets its own account

Never let the agent sign in as a person. When it has its own account:

The ERP system enforces the rules, not the connection

Between the agent and your ERP system sits a piece of connection software. It is tempting to write the safety rules there: “this agent must not read salaries.” Do not rely on that. A rule written in the connection can be undone by a mistake in the connection.

Your ERP system already has a permission system, built and tested over years. Use it. If the agent’s account cannot see salaries, no clever question will reveal them.

Read first, write much later

Reading cannot damage a record. Writing can. Start with reading only, and stay there until you trust what the agent reads and what it concludes from it. Letting it create an order or change a price is a separate decision, made later, one kind of action at a time.

How we did it

When we connected AI to our own ERP system at OdooTec, we made four choices:

The question to ask your team

“If the agent asked our ERP system for every salary, every bank account and every client’s price, which of those would it get?” If the answer is “we would need to check,” the access is too wide.