OpenText's VIM 26.2 notes say users can "Select and approve invoices via SAP Joule or any other agentic platform." That line usually starts a conversation about who clicks. We already covered who signs when an agent approves a VIM invoice: the Chart of Authority still names a person, the agent is a new route to the same button, and the approver should see the same flags they would see in the VIM workplace.
This piece starts one step earlier. Before the first agent click, who configures the channel that makes that click possible?
Configuration is a control. Someone turns the agentic path on, scopes which invoice types it may touch, binds it to identity and logging, and decides what still needs a human signature outside the agent. If that work has no owner, the first successful demo becomes the production design by accident.
In our view the pre-flight split looks like this. OpenText describes the capability; these role labels are ours:
- VIM process owner. Which approval steps, if any, may run through an agentic channel. Low-value, PO-matched invoices are not the same case as a non-PO invoice that just raised a bank-detail flag.
- SAP / BTP or Joule admin (or the equivalent for another agent platform). Which agents are allowed to call into VIM, under which technical user or principal, and how credentials rotate.
- Security / SoD. Whether the agent path preserves segregation of duties, and whether a shared technical account can never stand in for a named approver.
- Internal audit. What the VIM log must show: named approver, channel used, and enough context to tell a cockpit approval from a Joule one when someone samples next spring.
- Controller. What the human must still see or sign before money moves, even when the agent proposed the action.
What "any other agentic platform" forces. The blog does not limit the channel to Joule. That is flexible and easy to under-govern. Write an allow-list: which platforms, which environments (sandbox first), which company codes, which amount bands. A platform that is not on the list cannot approve. Silence is not an allow-list.
Human signature still means something. Agent selection and agent approval are not the same permission. Selecting a worklist item for review is a convenience. Approving spend is a delegated decision. Keep those as separate configuration choices. If your Chart of Authority requires a wet-ink or dual-control step for certain invoices, the agent channel does not erase it. Name where the human signature still happens: in VIM, in an attached workflow, or not at all for that invoice class.
Checks before the pilot. The public blog is one line. The release notes and your SAP security model carry the rest. Before anyone demos approval in chat:
- Identity. Does every agent approval resolve to a named person your Chart of Authority recognizes, or only to a technical user?
- Scope. Which document types and exception states are excluded until the channel is proven (bank-change holds, address-change holds, high-value non-PO)?
- Rollback. How the channel is switched off without stranding in-flight approvals.
- Evidence pack. One sample invoice approved in the cockpit and one through the agent, side by side, for audit.
We covered related 26.2 AI features elsewhere: document questions and the archive under them, plain-language process insights, and special handling for bank or address changes. Approval through an agent is the one that changes who may act. Configure the channel on purpose, then talk about who signs.
Where does your landscape stand?
Take the free Assessment. It's about 10 questions, and you get a one-pager with your estimated level of effort and suggested next steps.
