Tines Academy › Lesson

Prepare for team use

Agree who will maintain the agent, check access, and record what’s ready or still needs attention before team use.

Objective: Use test results and agreed responsibilities to decide what still needs to happen before team use.

Decide whether it’s ready for a team

You’ve reviewed the agent’s answers and available usage information. Those results are part of deciding whether it’s ready for a team. You’ll also need to check who can retrieve information and agree who will look after the agent.

This lesson brings that evidence and those decisions together. You don’t need to push the agent live to finish the path. If you’re building it to learn, keep it as a draft and note what would be needed for team use. You don’t need to move spaces, connect a real expense system, or involve other people just to complete the lesson.

Decide who looks after it

Choose an owner and a backup who can investigate problems and update the workflow. The owner doesn’t have to be the person who built it. Agree who looks after the policy and connectors, when to review answers and usage, how to update the policy, and how long to keep conversations and claim data.

Ask the policy owner to review representative answers, including conditions and exceptions. Also check who can inspect workflow storage and whether they’re allowed to see the saved information. Private conversations on the page may still be readable by people with storage access.

Agreeing when to remove saved data doesn’t make it happen. Before team use, that behavior needs to be added and tested.

You’ll record the owner, backup, and unfinished checks in Readme later in this lesson. The person who builds the agent may not maintain it, so make it clear who colleagues should contact.

Check access before team use

For practice, note unavailable checks in Readme and continue. You don’t need to recruit testers to complete this lesson.

Before team use, arrange two intended space members and one nonmember, using non-admin accounts. Each intended user needs a different approved test claim belonging to their account. The page is intended to use Space access, so the nonmember should be blocked.

Review the earlier access-test results

Read the results in workflow chat or Readme. Ask Tines 3B which earlier checks still apply to this draft and what remains to be tested with the additional accounts.

Check who can open the page

Have all three people open the same draft page link. The two intended users should get in; the nonmember should be blocked.

Check each user’s own claim

Each allowed user starts a new conversation and requests their own approved test claim. Complete individual connector sign-in if prompted.

Compare each answer with that user’s record. The agent should return the correct claim facts.

Request the other user’s claim

In a new conversation, each allowed user requests the other person’s approved test claim. Neither should receive details or confirmation that it exists.

Ask Tines 3B in workflow chat to show the lookup results too. The blocked record must stay out of the model’s input; a refusal in the final answer isn’t enough.

Check saved conversations

Have each allowed user create a conversation using made-up details. If the page provides conversation links, try the other person’s link. Ask Tines 3B in workflow chat to test direct requests for the other user’s saved history too.

Neither user should retrieve the other person’s conversation or history. Hiding it from the list isn’t enough. These page tests don’t replace the earlier review of who can inspect workflow storage.

Record the results

Ask Tines 3B to record the account roles, expected and observed results, and unfinished checks in Readme, without private content.

If something doesn’t match

Ask Tines 3B for the missing evidence or a focused fix, then repeat the affected checks before team use.

If accounts or draft access prevent a test, leave it unverified. Don’t widen access or grant admin rights to complete it.

Five go-live readiness questions

Record what’s ready

Review the earlier results and unfinished checks in the Readme against the five readiness questions. Send this prompt in workflow chat, stating whether you’re building for practice or team use.

Update this agent's Readme to explain what it does, which policy version it uses, its AI provider connector and model, and whether claims are fictional or come from an expense system. Ask whether this is a practice build or intended for team use if I haven't said.

Summarize the tests, known issues, operating limits, and checks still to do. Point to the test results behind each pass. Don't mark a check as passed if it hasn't been completed.

Record any agreed owner, backup, policy owner, and connector contacts. Include plans for reviewing answers and usage, updating the policy, and removing saved conversations and claim data.

For a practice build, mark roles and decisions that haven't been arranged as "Not yet agreed" without requiring me to resolve them now. For a build intended for team use, ask about missing decisions rather than inventing them.

Distinguish what we've agreed to do from what the workflow already does and we've tested. Leave out private claims, conversations, and credentials. Keep the workflow as a draft.

Check the readiness record

Read the update summary in workflow chat, then open Readme.

Check the recorded results

Compare the recorded checks with the test reports and what you observed. Each pass should point to a supporting result. Failed or unfinished checks should still be visible.

Check the responsibilities and decisions

Check that the owner, backup, policy owner, and connector contacts match what you agreed. The review arrangements, policy updates, and decisions about keeping or removing saved data should match too. Anything you haven’t decided should remain open.

Separate the plan from what’s working

The record should distinguish what you’ve agreed to do from what the build already does and you’ve tested. Writing down a plan doesn’t put it into practice.

If something doesn’t match

Ask Tines 3B in workflow chat to correct missing checks, passes without supporting results, or decisions you haven’t made. Reopen Readme to check the correction.

Keep practice builds as drafts, and finish the work needed for team use before inviting users.

Push live

Publishing is optional. If you’re practicing, still using fictional sources, or have important checks left to finish, keep the draft and continue to the knowledge check.

Review readiness

Check the readiness record and your organization’s required approvals, including the policy owner’s feedback and any corrections. Ask Tines 3B how the approved policy and settings will reach live without bringing over practice data or failure simulations.

Publish the draft

Select Push live in the branch selector. If asked about storage, you can choose to copy data from draft to live, or to start live with empty storage. If you’ve loaded your policy and claims into draft, choose to copy data to live.

Check the live answers

Open the live page from Links. Ask a known policy question, then request your own approved test claim in a new conversation. Complete individual connector sign-in if prompted.

Compare the answers with the approved policy and claim record. Check that the account connections work for the intended users.

Check access on live

Repeat the page-access, cross-user claim, and saved-conversation checks from Check access before team use, using the live page and approved test accounts. Include the direct requests for another user’s saved history. Finish these checks before the team starts using the agent.

If something doesn’t match

If someone can see information they shouldn’t, restrict access until it’s fixed.

Ask Tines 3B in workflow chat to investigate and fix the problem on a draft. After publishing the fix, repeat the affected live checks.

One last question about your own agent. At the end of the path, we’ll bring your answers together in a brief you can review and give Tines 3B to start building.

What to do if something still feels fuzzy

You worked through the readiness questions and recorded who’ll look after the agent or what still needs arranging. If you pushed it live, finish the live checks before the team starts using it. If you’re practicing, keep the draft and your unfinished checks in the Readme. Next, you’ll follow Leo through his first month live.

Build it safely: You checked which readiness decisions your results support and recorded what still needs testing or agreement before team use.