LAUNCH & LEARN · June 3, 2026
Give an agent one useful job.
Design a community assistant around a clear role, reliable sources and a result members can understand.
Chapters & searchable transcript ↓
Hosted by Ian McDonald
About 20–30 minutes to work through.
Leave with an agent brief you can test.
Before you start
Choose one small community task and a short set of approved source material. Start with a written draft rather than turning on unattended posting.
Use this guide as a practice companion. The recording captures the app and conversation as they were on June 3, 2026; historical announcements and screens may differ today.
Make one useful thing.
Choose the job
Write the audience, trigger and expected output. For example, summarize an approved discussion into a short recap. Keep the agent focused on a task you can evaluate.
Set the source and boundaries
List the sources it may use, the details it should omit and when it should ask for help. Give it its own clear identity instead of asking it to impersonate a member.
Test a draft response
Use a small example and inspect the result for factual accuracy, tone and privacy. Revise the brief before enabling any recurring activity in your community.
Give it your context.
These practice prompts adapt the session’s ideas. Replace the bracketed text, review the response and make one specific revision.
Write an agent brief
Draft instructions for a community assistant that [one task] for [audience]. It may use [approved sources]. Define its output, tone, limits and when to ask for human review. Give it a distinct identity.
Test the brief
Using these proposed instructions [brief] and this approved sample [sample], create one draft response. Then identify unsupported statements, private details and any instruction that needs clarification. Do not post it.
Finish with a next step.
0 of 5 complete. Your checklist is saved in this browser for this session.
Revisit the conversation.
Open recording (1231 MB) · Download transcript · Download captions
To save the video, use the player’s download menu.
This replay has been reviewed and trimmed where needed. Teaching, demonstrations and Q&A are retained. Chapters and captions follow the edited recording.
Jump to a chapter
Session transcript
Automatically transcribed from the recording and aligned to this edited replay. Wording and speaker labels may contain errors. Choose any timestamp to play that moment.
Loading transcript…
Ideas to listen for
A specific role beats a vague persona. The demonstration builds a community agent with an identifiable purpose and behavior.
Sources shape the answer. The group chooses recap-style updates. The live example uses a fictional placeholder transcript; your actual recap needs real, approved notes that you can check.
Get unstuck.
Does the agent need web research?
Only if the task requires it and you can review the sources. A recap of approved community material can use a narrower source set.
Can it pretend to be a member?
Give it a distinct assistant identity. Members should understand when they are reading an AI-generated response.
For help in your community, open Help → Support Agent and describe the result you want and where you got stuck.