DEV

When AI agents stop just reading ERP and finance data and start posting to the general ledger or releasing payment files, the security stakes change completely. This episode breaks down the four layers every team needs before going live.

Show Notes

AI agents that touch enterprise systems like SAP, NetSuite, and Salesforce are no longer just summarizing data — they're writing to the general ledger, updating vendor records, and releasing payment files without a human approving every action. This episode of DEV.co tackles the security architecture those deployments actually require, drawing from the detailed guide on securing AI agents for ERP, CRM, and financial systems. The stakes are high: a read error gives you a bad answer, but a write error gives you a misfiled journal entry, an erroneous refund, or a wire to the wrong vendor — and the auditor will want to know exactly who authorized it.

The episode walks through four layers of protection that enterprise teams can and should implement before any AI agent gets write access to a financial system:

  • Identity over prompts: Every AI agent needs a scoped, lifecycle-managed non-human identity — one service account per agent per environment, with roles limited to exactly what that agent needs. The agent's identity is the real control surface, not its system prompt.
  • Tool allowlists as policy: The model doesn't hold permissions — the tools do. Every write path should be a named, typed function with a server-side authorization check that runs independently of the model's decision, including amount ceilings, enumerated field values, and idempotency keys to prevent duplicate postings.
  • Credential hygiene via token brokers: Long-lived API keys shared across agent runs are the first thing auditors flag. A token broker authenticates the agent, mints a short-lived, audience-scoped OAuth token for the exact action requested, and ensures the invoking human's identity is carried as a claim into the ERP's own log — meaning dual attribution survives. DEV.co's cybersecurity development services address exactly these kinds of integration-level vulnerabilities.
  • SOX-compliant audit trails: Every financial posting triggered by an agent needs a structured, append-only log capturing agent identity, model version, policy version, full tool-call parameters, the authenticated human who initiated the session, and the downstream posting IDs returned by the ERP. Operational AI logs for SOX-relevant systems should be retained for at least a year.

A key theme throughout is that none of these layers require exotic technology — but most teams skip all four. The episode also highlights findings from the 2026 Verizon Data Breach report, which flagged machine accounts as the identities most likely to be abused in an agentic future, and OWASP guidance noting that a model choosing to call a tool is not the same as the user being authorized to use it. For teams building or extending these kinds of integrations, DEV.co's AI agent development services are designed to meet enterprise compliance requirements from the ground up.

Also worth a listen from the DEV.co feed: AI Has Killed White Label Web Development — What Survives explores how AI is reshaping another corner of the software industry.

DEV.co

RFP.co

What is DEV?

Software and web development from the side that has to ship it and then live with it. Architecture decisions with a cost attached, scoping, technical debt, hiring and vendor selection, and the AI tooling question every engineering team is now answering whether they planned to or not.

Each episode takes one decision — rewrite or refactor, framework choice, build versus buy, how to scope a fixed-bid project honestly — and works through the tradeoffs, including the ones that only show up in year two. Written for engineering leads, technical founders and the people who fund them. Five or six minutes, no hand-waving.

Topics include rewrite versus refactor, build versus buy, scoping fixed-bid work honestly, technical debt you should keep, framework and platform choices, hiring and vendor selection, code review culture, and where AI tooling actually helps.

Produced by DEV.co, web and software development. Full details, services and further reading at https://dev.co