AI & AutomationNov 19, 20266 min read

AI Agents in ERP: Why Governance Must Come Before Automation

Oracle NetSuite's Autonomous Suite illustrates where ERP systems are heading: AI agents that act autonomously — including write access to the general ledger. What this means for mid-sized companies and which governance questions must be answered before the first deployment.

AI Agents in ERP: Why Governance Must Come Before Automation — AI & Automation

ERP systems are the backbone of mid-sized businesses — accounting, inventory and order processing all depend on them. When AI agents inside these systems autonomously post entries, trigger purchase orders or adjust stock levels, that is no longer a distant future. Oracle NetSuite announced precisely these capabilities for its Autonomous Suite at SuiteWorld 2026, integrating industry-specific AI agents directly into its cloud ERP platform.

The critical question is not what agents are technically capable of — but what they are permitted to do. And who ensures they act only within the boundaries the business has deliberately approved.

What Oracle NetSuite's Autonomous Suite actually does

SuiteAgents are available for sectors including software companies, manufacturers, construction, healthcare and nonprofits. The offering is complemented by SuiteAgent Studio, where businesses can configure their own agents, schedule them and link them to business events.

  • SuiteSkills: over 45 embedded AI functions — from cash flow analysis to supplier assessment
  • AI Workspaces: central interface where humans, agents and data work together
  • Write access to the general ledger: agents can post entries directly — a new, highly sensitive capability
  • Approval workflows: actions require explicit user confirmation before they are executed

The governance problem: inherited permissions become an active risk

As useful as the automation is — it fundamentally changes the risk profile of the ERP system. A security analysis by Netwrix on the NetSuite AI Connector identifies the core problem: AI agents inherit the full permission scope of the connected account. What counts as over-permissioned for human users but rarely becomes a problem — because UI friction and habit act as filters — an agent can exhaust in milliseconds. Old roles and unreviewed access rights: latent risk turns into active action.

  • Prompt injection: hidden instructions embedded in PDFs, emails or database fields can redirect the agent — it then executes commands that did not come from the user
  • Hallucination: AI models can generate incorrect outputs and trigger erroneous bookings or purchase orders as a result
  • Permission inheritance: old, never-cleaned-up roles give the agent access to areas no longer needed — without anyone noticing

What mid-sized companies must clarify before the first agent deployment

Oracle NetSuite's framework demonstrates the right approach: agents operate within existing roles, permissions and approval workflows. The platform enforces technical boundaries — agents cannot make external HTTP requests or assume administrator rights. But the actual governance work lies with the business itself, before the first agent goes live. Companies that use structured IT consulting avoid the most common mistakes.

  • Least-privilege principle: create dedicated roles specifically for AI agents — with only the permissions the specific process actually requires
  • Access separation: never combine read access to financial data with write access to ledger entries in the same role
  • Draft mode first: agents start in review mode — they prepare actions, a human confirms. Full automation is activated only after proven stability
  • Audit logs: every agent action must be fully traceable — what was triggered, why, who approved
  • Regular access reviews: first access review no later than 90 days after go-live, then on a recurring schedule — to prevent permission debt from accumulating
AI agents turn existing permission errors active — what was previously latent risk becomes action.

What this means for mid-sized companies

Not every mid-sized company uses Oracle NetSuite. But the governance principle applies to any ERP system that offers or plans AI connectivity — from SAP to Microsoft Dynamics 365 to specialised industry solutions. The EU AI Act classifies systems that make automated decisions affecting finances, employees or supply chains as regulation-relevant. Governance is therefore not just best practice — it is increasingly a legal requirement.

Anyone integrating AI solutions into existing enterprise software today does well to ask the governance questions before the first agent goes live — not after. For a practical look at technical and organisational preparation before AI deployment, see our guide Before AI Gets Started.

AI integration for mid-sized companies — what we assess before agents go live

This article was created with AI assistance and editorially reviewed.

Have an idea worth building?

Tell us where you want to go. We'll help you get there with software that performs.