In the first week of July, two things happened simultaneously, both related to how ERP manages AI.
One is that Workday released a complete set of products at DevCon 2026: Agent Passport (issuing passports to AI agents), Developer Agent (allowing developers to build agents in the IDE using natural language), and Agent-Ready Tools (enabling agents to directly operate HR and financial data while inheriting Workday's permission model).
Another is SAP. A report from German media Handelsblatt says that SAP is tightening its hiring and travel rules — not because it lacks money, but to free up funds to invest heavily in AI. CEO Christian Klein’s original words: "We are not losing contracts because of AI; we are winning contracts because of AI."
Two things convey the same principle: ERP vendors are no longer just piling on AI as features. They are re-architecting the entire product, making AI Agent a native layer of ERP. And governance—who manages these Agents and how—is becoming the core of competition.
Agent Passport: A Strategic Inflection Point You May Not Notice
I carefully read through this wave of Workday product releases. The one that most easily catches attention is Developer Agent — an AI application for natural language development, which sounds very flashy. But I believe the one with the most long-term value is Agent Passport.
What Agent Passport does sounds simple: testing, verifying, and continuously monitoring every AI Agent—whether built by Workday itself, by customers, or by third-party vendors. The testing standards are tied to three public frameworks: OWASP LLM Top 10 (large language model security), NIST AI RMF (AI risk management), and MITRE ATLAS (adversarial attacks). Cisco is the first certified partner.
What does this mean?
I'll give you a scenario. Suppose you're using Workday to manage payroll. A third-party AI Agent wants to access your payroll data — can it read? Can it write? Has it undergone security testing? If it gets compromised, who is responsible?
Before Agent Passport, the answers to these questions required you to check each vendor's documentation one by one. Now Workday provides a structured answer: which items the Agent has been tested on, who tested it, which standard it is benchmarked against, and when the validity expires. These items have a signable audit trail.
"Agent trust will become a buying conversation."
—— ERP Today Analysis
Agent trust is becoming a procurement issue. When I chatted with a few CIO friends at the end of 2025, they hadn't taken it seriously yet. Half a year later, Workday has already productized it.
SAP's "Cost-Saving" AI Investment
SAP's approach differs from Workday's, but the direction is the same.
Klein is tightening budgets—stricter hiring, leaner travel. This isn't SAP in decline; it's SAP making strategic trade-offs. The organizational restructuring initiated 18 months ago has now entered deep waters: cutting slow-growth positions and shifting resources to AI and fast-growing business lines.
This logic is not hard to understand. Investors in the software industry are all asking one question: Will AI enable companies to build their own tools, thereby bypassing traditional software vendors? Klein has been pushing back against this narrative, but his actions show that he at least takes this risk very seriously.
From a data perspective, SAP hasn't played its AI cards too poorly. The over 200 AI agents and the Joule unified interface launched at Sapphire 2026 have already seen real customer adoption. But the question is: when the entire industry is doubling down on AI, is your investment fast enough?
Klein's answer was straightforward: cut all non-core expenses and throw money and people at AI.
Forbes's Cold Water: Agents Overwhelm Infrastructure Before Delivering Returns
Just as ERP vendors were busy launching products and adjusting budgets, Forbes published an analysis in early July. The viewpoint is not particularly novel but is very practical: many companies are deploying Agentic AI as an "upgraded chatbot," but when scaling up, these agents will bring down both infrastructure and budgets.
Forbes' core viewpoint:
Proactive agents continuously monitor conditions, make decisions, call APIs, and trigger thousands of small-scale context-rich interactions. Moving from demonstration to production without a robust platform will overwhelm existing infrastructure and budgets.
—— Core recommendation: Define an internal "Agent platform" that includes central routing, observability, and cost control before approving widespread deployment.
This reminder is especially critical in the context of ERP. ERP is different from other software—it manages orders, inventory, finance, and payroll. When an AI Agent makes a mistake here, it's not the kind of error like "pushing irrelevant content to users," but the kind like "overpaying 100 people's salaries."
So when Workday launched Agent Passport, it wasn't just selling a security feature. It was answering a question that every CIO is bound to ask: "How do I know this agent won't mess up my core business processes?"
Similarly, when Odoo makes the MCP server native, its logic is the same—not to let AI bypass the ERP permission system, but to inherit it.
New Dimension of ERP Selection: Governance Capability
Let me put it bluntly: In the past, ERP selection focused on functions, price, and industry experience. In the second half of 2026, ERP selection will have an additional dimension — governance capability.
| Selection dimension | Traditional ERP | ERP in the AI Era |
|---|---|---|
| Functional Coverage | Whether the module is complete | Can AI Agent operate core processes |
| Integration Capability | Is the API open | Whether it natively supports Agent protocols such as MCP |
| Safety and Governance | RBAC permission model | Agent testing, certification, monitoring, and revocation full chain |
| Scalability | Development Platform/Low-Code | Natural language development Agent + automatic governance inheritance |
| Ecology | Implementation Partner | MCP connection capability of partner + AI ecosystem |
These five dimensions were not casually listed. Workday's current product covers three dimensions: integration (MCP+Agent-Ready Tools), governance (Agent Passport), and expansion (Developer Agent). Odoo's MCP server covers two dimensions: integration and ecosystem. SAP's budget adjustment essentially strengthens its weakest governance and AI capabilities.
In other words: the competition among ERP vendors is no longer about "who has more modules," but rather "who can make AI agents operate safely."
Four Practical Suggestions for CIOs
Having covered the manufacturers, let's return to your account.
First, you can start evaluating whether your ERP vendor has an "Agent governance" solution right now. Regardless of whether you are using Workday, SAP, Odoo, or a domestic ERP, ask one question: Does your AI Agent have a certification system? Can the Agent's operations be audited? Can an Agent's permissions be revoked at any time? If the answer is no, note down this gap.
Second, don't wait until the selection phase to discuss governance.If your company is already using AI agents—even if just a few pilots—first take stock of them. Which agents are operating which systems? What data have they accessed? Have they been tested? Is there human oversight? This inventory process alone can uncover a host of issues.
Third, the MCP protocol deserves serious attention. Workday uses it, Odoo uses it, and Anthropic, OpenAI, Google, and Microsoft all support it. MCP is becoming the universal interface for ERP-AI integration. If you are planning AI-ERP integration, prioritize products that support MCP to reduce the workload of custom integration.
Fourth, pay attention to infrastructure costs.Forbes’ reminder is not an exaggeration. One Agent may have little cost, but 100 Agents calling ERP APIs, reading data, and generating operations every day — these costs add up, and if not controlled, they will accumulate much faster than you think. It is recommended to set up cost monitoring and usage limits before deployment.
