Ledgers and work orders begin to move on their own: Before enterprises hand "execution authority" to AI, they must first hold two lines.

Enterprise Digitalization Observation ｜ 2026-11-02 (Issue 2) ｜ Perspective: Execution Authority and Governance Boundaries

At the end of October, two things happened that are worth every person managing digitalization pausing to think about. One: a company called Serval raised $150 million from Sequoia and saw its valuation surge to $3 billion; what it does is very straightforward — it lets AI write automation itself. The other: in the finance field, AI bookkeeping tools are making “review first, then post” an industry standard. One is seizing execution authority, while the other is guarding review authority. Viewed together, these two things explain better than any vendor white paper: when AI enters companies’ core processes, the real contest is not “whether it can do the job,” but “who has the final say.”

1. When AI Starts Writing Its Own Work Orders

Serval's model isn't complicated: employees describe a workflow in plain language, and AI writes the automation, then before going live it goes to an administrator for review. It sounds like an evolved version of a low-code platform, but its strength lies in this—it also automates the very act of "writing automation." In the past you had to maintain an implementation team to build processes; now AI drafts first, and people just review.

Even more striking is the counterattack from its benchmark target: ServiceNow launched the standalone AI service desk Flow on October 1, charging "$2 per conversation." On one side, AI is collapsing workflow costs and turning services-heavy software categories into products; on the other, incumbents are using usage-based pricing to catch this wave of demand. This shows one thing: The "cost collapse" of agentic execution has truly happened, and work such as ITSM, ticketing, and process automation—where people write rules and people run processes—is being repriced.

II. Finance, however, goes the opposite way: first strengthen 'review'

But even as AI enters core processes, finance takes a different path. AI bookkeeping tools represented by Truewind do not offer "automatic posting," but rather "voucher suggestions with an evidence chain"—it extracts invoices and receipts, reconciles them against your existing chart of accounts, and generates a reviewable journal entry suggestion, and posting occurs only after accounting review approval. Odoo Accounting follows a similar approach: sales orders and purchase orders are uniformly posted to the accounting module; AI speeds up matching and reconciliation, but does not make the final call for you.

III. An open-source foundation keeps "governance rights" in one's own hands

This brings us to the unique value of open-source ERPs like Odoo. When an agent runs in your system, the most dangerous thing is not how fast it works, but that you cannot clearly see what it has changed. Closed-source SaaS puts the three things—configuration, review, and audit trails—into its own black box, and you can only trust the reports it gives you; Odoo 20, by contrast, returns API and data sovereignty to you—self-hosting, full database copies, and traceable audits. At the very least, you can see clearly: which record this agent changed, who configured it, and what traces it left.

Open source is not a byword for "cheap"; open source is insurance that "governance rights will not be held hostage." When agents start executing on their own, this key should be kept in one's own hands.

Conclusion: Execution can be outsourced, but accountability cannot be left hanging

Delegating execution authority to agents is the general trend and the most direct lever for cost reduction. But remember two lines: First, high-stakes actions must never be executed autonomously—when funds, compliance, or account closing are involved, there must be human review and traceable logs; Second, governance authority must remain in systems that are autonomous and controllable—who does the configuration, who signs off on reviews, and how traces are kept: these three things cannot be outsourced to a black box you cannot understand. The stronger the technology, the more clearly these two lines must be drawn. The ledger may move on its own, but responsibility must be taken by someone.

This article was originally written by "Haoyunyun" Enterprise Digitalization Observation | 2026-11-02 (Issue 2)

Reference sources: The FundForge / SiliconANGLE / Fortune (Serval, 2026-10-29), gaugius's '2026 Best AI Bookkeeping Software', and iVentureTeam's 'Odoo vs Dynamics 365 BC'. This article is for industry research reference only and does not constitute product selection or investment advice.