Your China-based content partner has moved the production workflow into a different AI assistant. The asset library connects and the tools respond. Before handing over the next launch, ask a different question: will the new setup still apply the review decisions your team agreed?

ONCE recommends accepting connectivity and workflow behavior separately. Carry a small set of comparison briefs with explicit expected outcomes into the new environment. What should it produce, what information should it flag, and which approval should it wait for? These are more useful acceptance criteria than identical wording from two assistants.

A recent integration makes the distinction timely

Adobe’s November 4, 2026 announcement says CX Enterprise Coworker is available in ChatGPT, Claude, Databricks Genie, Gemini for business and Microsoft Copilot. It also introduces MCP connectors and describes agencies customizing services with proprietary IP, tools and expertise.

The announcement establishes Adobe’s stated integrations and partnerships. It does not establish identical configuration or workflow performance for every customer. ONCE’s inference is that procurement should increasingly examine the business process carried through these entry points, alongside the connections themselves.

A shared protocol does not validate an identical process

The official MCP blog’s November 3, 2025 article on server instructions explains that hosts determine how they use those instructions; system-prompt injection is not guaranteed. It recommends evaluating client behavior and warns that written instructions cannot guarantee consistent model behavior.

This is not evidence that Adobe or another specific product skips reviews. It is a reason to verify the actual environment. Brand inputs, workflow guidance, available tools and platform settings all matter. If a particular approval must precede execution, ask the supplier to show how that dependency is implemented rather than pointing only to a sentence in a prompt.

Write expected outcomes before running comparison briefs

Start with a narrow representative job: turning approved product material into a reviewable video script. Use anonymized or fictional inputs in a trial environment that cannot publish or alter live business systems. ONCE proposes three starting cases, not a complete reliability test:

  • Complete material: produce the requested draft, identify the reference version and wait for the agreed approval.
  • A missing supporting document: flag the unsupported claim instead of borrowing a number from an older example.
  • A changed market or bundle: identify approvals that no longer apply rather than translating the old offer unchanged.

Define the expected outcome before seeing the answer. Imagine a fictional overseas appliance brand sourcing in China. Its previous script includes an accessory supplied only in a bundle. The new brief is for the standalone item. A useful result preserves supported product benefits while addressing the changed accessory statement. Rejecting the whole assignment or copying the old script does not complete that job.

Compare business decisions and usable deliverables

Retain the input, output, reference versions and available records of platform, model, tools and settings. Check approved facts, necessary qualifications, review stops and handoff files. Different prose can be acceptable. Dropping an important limitation is not merely a style change. A statement that the assistant “checked everything” does not replace examining what it delivered.

Repeat important or inconsistent cases and retain differing results. One successful run supports that sample and configuration, not a permanent reliability rate. When a case fails, identify whether material was missing, guidance was not used, a tool behaved differently or a setting needs adjustment. Do not rewrite the entire brand knowledge base before locating the problem.

If the destination cannot perform an existing step, document a retained human task or original tool. A workable migration can include explicit exceptions. It should not disguise unfinished steps as complete automation.

Purchase a transferable method with stated dependencies

Specify three handover items: approved brand inputs with versions; readable workflow steps, review points and exception handling; and acceptance samples with actual results. Document platform-specific connections, permissions and capabilities that cannot simply be exported. Agree the permitted use of client material and supplier-owned methods. An open connection standard does not settle ownership or make every capability portable.

Separate input preparation, workflow configuration, migration trials and maintenance in the scope. Reuse business rules that remain approved, then address platform differences. If the product or market also changes, recognize the additional business review involved. Explore ONCE’s video marketing articles for related content planning.

For ONCE’s move toward managed creative services, the potential asset is a reusable set of business examples, expected outcomes and revision reasons. Its labor and commercial value must be established in actual projects, not inferred from the number of connected tools.

Scope: Primary sources checked November 8, 2026. The comparison method and appliance example are ONCE recommendations and hypothetical planning, not client results or tests of Adobe software. The original AI cover is conceptual.