Your next AI assistant should inherit what your team taught the last one. Before buying another platform, ask whether your working instructions, examples, and review standards can move with you. Then test the answer on a real assignment. Your business should retain the benefits of repeated AI use when a subscription changes.

Anthropic announced on October 22 that it is donating Agent Skills to the Agentic AI Foundation under the Linux Foundation. The company says more than 70 agent platforms have integrated the standard since December 2025, including ChatGPT, Cursor, GitHub Copilot, and VS Code.

That is a reason to ask harder buying questions. It is not proof that your workflow will run unchanged everywhere.

In brief

Keep the playbook your team develops through AI use. Choose one recurring assignment, capture the instructions and corrections that make it work, and test them with a second compatible agent. Measure how much rebriefing and repair remain. Waiting six months costs more when every useful lesson stays inside a private conversation.

The asset is the method you worked out

Most organizations can name the AI products they pay for. Ask what they have learned to do reliably with those products, and the answer gets less precise.

Someone in finance has a conversation that produces a useful variance report. A sales manager knows how to get a decent account brief. An operations lead has finally taught an assistant which exceptions belong in the weekly update.

Those are working methods. But if the instructions live in individual chat histories, the company may not be able to reuse them when a colleague takes over. Switching platforms exposes the same weakness.

The Agent Skills project describes its format as a folder containing instructions, with optional scripts, reference materials, and templates. Its stated purpose includes packaging company-specific procedural knowledge for reuse across compatible agents. In business terms, that is a way to give an assistant the playbook for an assignment without dictating it from scratch every time.

The folder is easy enough to copy. What makes it worth keeping is the correction someone adds after a brief misses a customer commitment.

Consider a hypothetical account-review brief. The first version summarizes everything. A manager corrects it: unresolved customer commitments matter more than activity counts. The next version misses a renewal clause. That becomes a required source. Later, someone adds an example showing how to separate confirmed facts from sales assumptions.

Save those changes as you go, and the next person can use what the team has already worked out. Leave them in the chat, and that person starts over.

Portability has two different meanings

A vendor can truthfully say it supports Agent Skills while your assignment still needs substantial repair after a move.

The instructions might transfer. Access to your customer records might not. One platform may have the necessary document tools; another may require a different connection. A script may depend on software the destination does not provide. Different agents may interpret an ambiguous instruction differently.

A portable playbook requires a separate permission decision in its new environment.

This is why I would separate two questions in a buying conversation. Can we retrieve the working materials in a readable format? And can another approved system use those materials to complete the assignment to our standard?

I would want to see both before signing.

Simon Willison’s October 22 coverage points to the donation as a move toward standards-body stewardship. That is encouraging for a shared format. It does not establish equivalent results across products, or make every downloadable skill safe to run. Those remain your team’s testing and security responsibilities.

Ask for a working handover

For the next evaluation, bring a small assignment your team already understands. Use approved, non-sensitive examples. A weekly account brief is enough to test the handover with limited access.

Give the first agent the current instructions, a sample of an acceptable result, and the sources it may use. Have the person who normally reviews the brief correct its output. Save the correction in the working instructions, then rerun the assignment on another example.

Now ask the vendor to show the handover. Retrieve the instructions and supporting materials. Give them to a second compatible agent with separately approved access to the same test sources. Keep customer messages and record changes outside the trial.

Watch where the second agent struggles. Did an important rule never leave the original conversation? Is a reference missing? Does the process depend on an unavailable tool? Or was the instruction unclear enough that the first agent only appeared to understand it?

Record the review time and the missed requirements, then compare those with starting from an empty conversation. If the imported playbook still takes an afternoon of rebriefing, put that cost in the evaluation.

Six months should leave you with something

Six months from now, someone else may have to produce that account brief. They should be able to find the required sources, the manager’s corrections, and an example they can trust without asking who has the original chat.

That advantage is earned by capturing working methods as the team uses a compatible platform.

For an agentic organization, the purchasing standard should be straightforward: the business keeps the methods it develops. When the assistant changes, the next one should inherit a better starting point.

Research and structure: Mai. Editorial direction and voice: John Lipe.