WEBVTT

C01
00:00:03.670 --> 00:00:09.855
Open Lab, choose New agent, and start the guided creation flow.

C02
00:00:09.575 --> 00:00:17.817
Choose From prompt to describe the Agent directly, or From chats to include examples of the work it should handle.

C04
00:00:17.537 --> 00:00:27.159
This synthetic example starts with three chats: Acme has renewal blockers, Northstar is expanding, and the team needs a weekly account review. Add them with the Agent purpose.

C06
00:00:26.879 --> 00:00:36.819
Review the selected excerpts as examples. OpenPond turns their recurring goals, decisions, corrections, and answer patterns into the first Agent plan.

C07
00:00:36.539 --> 00:00:46.658
The draft converts those examples into the Agent's purpose, chat behavior, runnable actions, output formats, and Evals for your review.

C08
00:00:46.378 --> 00:00:52.652
The right side panel asks how the Agent should prioritize conflicting account risks.

C09
00:00:52.372 --> 00:01:00.753
Before creation, review the proposed name, chat endpoint, runnable actions, Evals, and output formats.

C09A
00:01:00.473 --> 00:01:09.071
If the plan misses a requirement, open Edit plan and describe the change in plain language before any Agent source is created.

C09B
00:01:08.791 --> 00:01:16.049
Review the revised plan, then click Confirm plan to create the Agent and run its Evals.

C10
00:01:15.769 --> 00:01:24.189
This is the creation result: OpenPond built the Agent and passed its behavior Evals against the Account Health examples.

C11
00:01:23.909 --> 00:01:31.346
After the Evals pass, the Agent is saved to your Profile and is ready to use.

C12
00:01:31.066 --> 00:01:37.658
The Agent page shows its purpose, chat entrypoint, and runnable account actions.

C13Q
00:01:38.548 --> 00:01:46.929
Use opens a normal chat with the Agent attached. Type a natural-language question about Acme before sending it.

C13
00:01:46.649 --> 00:01:56.410
Acme is the at-risk example: the Agent answers in chat with renewal timing, usage decline, the disputed invoice, and the open P1 case.

C14
00:01:56.130 --> 00:02:04.728
A follow-up keeps Acme context and assigns the first actions to Revenue Operations and Support instead of restarting the conversation.

C15
00:02:04.448 --> 00:02:13.086
The action picker offers chat plus each runnable account action, so you can choose conversation or a focused task.

C16
00:02:12.806 --> 00:02:23.371
Northstar is the healthy expansion example from the source chats. Its direct summary identifies rising usage and the request for twenty-five more seats.

C17
00:02:23.091 --> 00:02:31.650
Running renewal triage for Acme returns high risk, cites the billing and support blockers, and orders the next steps.

C18
00:02:31.370 --> 00:02:40.188
The weekly review action combines the portfolio and returns visible Markdown, CSV, and JSON artifacts for downstream use.

C19
00:02:39.908 --> 00:02:47.216
After restarting OpenPond, the same Agent, chat history, and actions remain available.

I00
00:02:48.106 --> 00:02:58.800
The current Agent finds the right facts, but buries the priority after the account summary. Improve will make the billing and P1 decision lead every high-risk result.

I01
00:02:58.520 --> 00:03:08.143
Open Improve agent on the existing Agent, then choose whether to describe the correction directly or include a chat that demonstrates it.

I02
00:03:07.863 --> 00:03:18.289
Choose From chats, ask every high-risk response to lead with the billing and P1 decision, and attach the correction conversation that demonstrates it.

I03
00:03:18.009 --> 00:03:27.591
Review the new lead-with-priority behavior, the existing capabilities that must stay intact, and the Evals OpenPond will rerun.

I04
00:03:27.311 --> 00:03:33.904
Review the planned change and confirm that it preserves citations and every existing action.

I05
00:03:33.624 --> 00:03:43.335
The same locked Evals make the decision clear: the current Agent misses the new lead sentence, while the updated Agent passes every original and new Eval.

I06
00:03:43.055 --> 00:03:52.102
Apply the update only after that comparison. The existing Agent updates in place, with the successful Evals recorded.

I07
00:03:51.822 --> 00:04:01.314
The improved chat now places disputed billing and the open P1 case ahead of adoption decline while preserving source citations.

I08
00:04:01.034 --> 00:04:09.504
The direct triage action repeats the billing-first result, proving the change works outside chat without breaking the existing workflow.

I09
00:04:09.224 --> 00:04:17.376
Commit the final Agent source with a clear Git message so the exact Profile version remains reviewable and shareable.

I10
00:04:17.096 --> 00:04:30.334
Open Sync to review the committed Profile changes. Confirm sync publishes this version so teammates can use the Agent in chat and hosted sandboxes can attach the same source.
