Give AI a speaking role in your meetings.
Give an AI a job in the conversation. It might confirm a new to-do while everyone can correct the owner, keep track of decisions, suggest an idea when the group gets stuck, or check a claim against the documents. A more ambitious agent can arrive with a mission and reference material, then ask to speak when it has something useful to add.
Try it now
Copy the prompt into your AI tool of choice, then answer its questions as they come.
Paste it into any AI chat to start. For most people, this is all you need.
Everything on this page in one paste: the prompt plus the full plan, every step and what to check. Use it for longer moves, to plan ahead, to share with a colleague, or if your AI can't open links.
How to run it
Choose one job for the AI in an upcoming meeting. Start with the contribution you want: confirming to-dos, recording key decisions, offering ideas, or researching facts. For a specialist role, choose a concrete mission such as checking proposals against the launch brief. The prompt below can investigate your company's tools and guide the setup through the meeting.
These are general instructions, not a description of this user's particular AI system, tools, access, or circumstances. Adapt the method to the user's goal and the capabilities you can actually use, while preserving the activity's essential sequence, review points, and intended result. ### Start with the activity and the user's context Read the complete activity and guide the user through it. Use the activity prompt yourself; the user does not need to paste it again. Keep the responsibilities of the user, the AI, and any colleague clear. Begin with what the user has already supplied. Establish the intended result and resolve missing context through focused questions when needed. Work within the activity's stated scope, permissions, required confirmations, and stopping points, including any deliberate order for introducing information. Make progress with the available information and reasonable, clearly identified assumptions where the activity allows them. ### Check the capabilities needed for this work Inspect relevant available tools, connectors, and data. Use small, relevant, non-destructive checks to establish whether necessary access or a capability actually works. Naming a product does not establish access to it. Let this activity determine what needs checking; do not impose a universal preliminary questionnaire or a full system audit. When useful access is missing: - Explain what you cannot reach and how that material would help. - Investigate practical setup, connection, export, or other access alternatives before asking the user to gather material or troubleshoot. - If progress depends on access, request only the specific user action or permission needed to enable it or provide the material another way. Continue useful independent work while awaiting the response. - Offer to walk the user through an export or upload. Welcome files, links, copied text, rough notes, or an interview as supported by your system and the activity, and use the best available route. If the missing access is optional, continue with what is available. ### Ask focused questions and respect the answers Ask questions only when necessary to complete the work well or satisfy the activity's required review points. Build on answers already supplied. Ask one question at a time, or a small group of closely related questions. Wait on work that depends on the answer, and continue useful work that does not. Explain unfamiliar language with examples where helpful. Confirm answers or decisions you supply on the user's behalf before treating them as the user's final answers or decisions. This does not require separate approval of every sourced finding; follow the activity's specific review requirements. ### Keep evidence and independent contributions distinct Distinguish evidence, inference, and missing information. Check your work against the source material. Use independent agents when your system actually supports delegation and independent investigation or review would improve the work: - Give each agent a bounded brief and the context appropriate to its role. - Keep discovery independent until synthesis where the activity requires it. Preserve any requirement for the user or colleagues to contribute before seeing other explanations. - Follow the activity's stated fallback and any required confirmation when independent agents or fresh contexts are unavailable. Preserve reviewer isolation where required; describe shared-context or sequential passes accurately. ### Produce, check, and revise the result Do the substantive investigation and production. Carry the activity through to the most complete, usable result within its scope, rather than ending with a plan or handing the main work back to the user. When missing information does not prevent useful work, proceed with clearly identified assumptions where the activity permits them. Check the result against the activity's completion criteria. At each substantive delivery, state what is complete and any material unfinished work, assumptions, helpful missing information, and remaining uncertainty. Explain how these affect what the user can use, and help the user revise the result. Preserve the activity's stopping point. Do not turn a rehearsal, investigation, or test-design activity into execution that the user has not authorized. Help me give an AI a useful, audible role in a real meeting. This is a fast-moving area: first find what is available in my company and what current commercial or open-source tools could make it work. Then help me configure one bounded role, test it, and use it in a meeting. Do not stop at a list of tools. ### Find the meeting context and our tools Use the meeting link, project, or rough description I have supplied. Search accessible calendar entries, agendas, relevant email and chat threads, earlier meeting transcripts, and project documents to establish the meeting's purpose, participants, outstanding decisions, existing commitments, and useful reference material. Look for participants' actual questions and concerns. Keep documented positions separate from inferred ones. Search accessible company software catalogs, IT guidance, license records, approved-app lists, and prior support messages for the meeting platform, AI assistants, bot/app permissions, and existing research or document connectors. Actually inspect available access. If you cannot inspect an account or connect to the meeting, say exactly what is missing and investigate the setup path before asking me to do anything. Ask one focused question at a time for missing choices: the meeting, the job the AI should perform, whether it needs to speak directly, and who should hear or see its contributions. Use context already supplied; do not make me prepare a document packet first. ### Check the current product capabilities Search current primary documentation, official app listings, and source repositories. Date the findings and distinguish shipping functionality, account-gated rollout, preview, demonstration, and custom development. Check our existing platform's options first, then compare the relevant commercial products and at least one credible open-source route if a custom role would benefit us. Starting references to recheck: - Teams Facilitator: https://support.microsoft.com/en-us/teams/platform/frequently-asked-questions-about-agents-in-microsoft-teams - Otter voice interaction in Zoom: https://help.otter.ai/hc/en-us/articles/30839137508631-Otter-AI-Chat-with-voice - Otter voice actions: https://help.otter.ai/hc/en-us/articles/30910450353431-Otter-AI-Chat-voice-commands-and-actions - Otter Live Assist and Agent Builder: https://otter.ai/blog/otter-ai-introduces-live-assist-the-first-live-coaching-agent-for-every-call - Attendee source: https://github.com/attendee-labs/attendee - Attendee speaking-agent example: https://github.com/attendee-labs/voice-agent-example - Attendee media capabilities: https://attendee.dev/changelog - Fireflies Voice Agents: https://guide.fireflies.ai/articles/6333219586-how-to-view-your-voice-agent-meetings Verify the functions independently: - Hear the current meeting through live audio or a live transcript. - Generate responses while the meeting is happening. - Speak audibly to everyone, answer voice commands, or only display private/chat responses. - Read the reference documents and search sources needed for this specific role. - Display an action item with its owner, and support corrections while the group is present. - Wait for a cue, raise a hand/request a turn, or interrupt under specified conditions. - Join this exact platform and meeting type using our account and device setup. A product that transcribes meetings or hosts its own AI interview does not necessarily join our meeting as a speaking collaborator. An MCP connection to past transcripts does not establish live listening. A generic voice model does not establish meeting integration. Verify the whole path. Identify licenses, rollout status, who must install/enable/admit the app, where its audio and visible output appear, current cost when established, and the work required to create a custom integration. Do not assume open source means no hosting, speech-model, or operating costs. Give me a short comparison and recommend a route for my chosen role. If no direct speaking route is available, propose a clearly labeled human-hosted version: an approved AI receives selected context or available live transcript, and a person reads its contribution aloud or plays approved audio. Explain what that version cannot monitor or do on its own. Wait for my choice if the fallback changes the intended activity. ### Create one useful role Begin with one role; offer more ambitious behavior only if it serves the meeting. Write a ready-to-use agent brief containing the mission, reference material with dates/source links, available tools, expected output, speaking rules, and completion point. Configure those settings where the product permits it. Where it does not accept custom instructions, turn the brief into the exact supported settings and commands; do not imply pasting a prompt changes a fixed product's behavior. Possible roles: - Action recorder: capture a proposed task, named owner, due date only if supplied, and a short supporting quote/time. Show it while people can correct it. Ask for owner/date clarification only when needed; do not invent assignments. - Decision recorder: separate an agreed decision from a suggestion or unresolved debate. Record who confirmed it, its scope, and the relevant source/time; bring back ambiguous items for the group to settle. - Idea generator: use the actual meeting question and constraints to offer a few distinct ideas, explain what is new about each, and mark them as proposals. Contribute on invitation or at an agreed agenda point. - Fact checker/researcher: search the supplied company record and approved external primary sources for claims that matter to the decision. Show the exact claim, what the sources support or contradict, their dates, and unresolved uncertainty. Separate checking against supplied material from doing a live external search. - A specialist with a mission: for example, compare proposals against an agreed launch brief or customer requirements. Ground the mission in actual documents and define the conditions that justify seeking a turn. Preserve this possible simple design: when the AI captures a to-do, it makes a little sound and displays the task and assignee so the group can confirm or correct them. Check whether the selected tool actually does that. If it supports the task display but not the sound, state that precisely; offer an audible spoken confirmation or a small custom addition only if I want it. Do not attribute an unverified chime feature to a vendor. Set a speaking policy with me. It can be invitation-only, at named agenda points, or allowed to request a turn when a specific trigger occurs. For an interrupting agent, define those triggers, how it waits for a natural pause, how brief it should be, and how the host can silence it. Do not convert permission to speak into permission to assign people work, send messages, or change company systems. ### Set up the chosen route Prepare the role brief and load the relevant reference material within the allowed tool. Verify that the agent can actually retrieve a named fact from the material and identify its source. For an available commercial agent, do permitted configuration yourself, then give the host only the required enablement, joining, and audio actions you cannot do. Prepare a short introduction so participants know the agent's role and how to correct or pause it. For a custom route, produce an implementation brief for the existing technical owner or coding assistant: live meeting input, reference retrieval, model/tool calls, turn-taking, speech output, visible task/decision output, correction storage, and a host stop control. Specify which parts already exist in the selected project and which must be built. If implementation is authorized and you have the capability, build the smallest version and verify it. Otherwise deliver the exact bounded handoff and its acceptance test; do not call a repository link or untested demo a working agent. For the human-hosted version, prepare short prompts and an easy way for the host to provide the selected context and bring the result into the room. Keep the substantive research and drafting with the AI. Do not require a person to secretly transcribe the whole meeting or pretend the AI heard what was never supplied. Ask for any necessary install, purchase, external connection, or actual bot-invitation authorization before taking that action. Prepare the concrete configuration first. Keep access requests distinct from the finished meeting outcome. ### Rehearse the role end to end Prepare a short realistic rehearsal using the same accounts, platform, and output channel as the real meeting. Confirm everyone intended to hear the AI can hear it, and that visible output reaches the intended people. Check that the system does not listen to its own output and repeatedly respond. Exercise the chosen role: - For tasks/decisions, include one genuine commitment, one hypothetical suggestion, a corrected assignee, and an unspecified deadline. Check the displayed record after the correction. - For ideas, pose an actual decision question, then add a constraint and check whether its suggestion changes. - For research, use one supported claim, one incorrect claim, and one claim the provided sources cannot settle. Check citations, source dates, and its willingness to leave the last one unresolved. - If proactive speech is enabled, demonstrate an allowed trigger, ordinary discussion where it should stay quiet, a person speaking over it, and the host's pause/stop command. Observe results directly when possible; otherwise ask for the specific missing evidence. Diagnose failures and repeat the affected test after a fix. Do not claim it is ready because configuration saved successfully. If the route cannot perform its speaking role, agree a reduced role or fallback before the real meeting. ### Use it in a meeting and evaluate the contribution Support the actual meeting only through capabilities you have verified. Make clear whether you are connected live or helping before/after. At the end, collect the confirmed tasks or decisions, proposed ideas, or sourced research results appropriate to the role. Ask the group whether the AI contributed something useful and whether its timing helped. Record one or two concrete examples, corrections, and the change to make before the next meeting. Leave an updated reusable role brief and setup instructions. Mark completion only after a real meeting includes the chosen audible AI contribution, either directly or through the explicitly agreed human-hosted version. An implementation brief, silent recorder, or post-meeting summary alone is a partial result. These instructions were written in September 2026 using the frontier models available at the time. You may now have better tools or models, or restrictions that make some tasks harder. Your AI should adapt to the tools and access you actually have and find the best available way to help. Everything here is a starting suggestion for you and your AI, not a proven best practice for every setup. Feel free to rewrite any of it to better meet your needs.
Find a route that can do that job during your actual meeting. A note-taking bot, a private coaching panel, and a participant that speaks to everyone are different experiences.
Give the agent a short brief: its mission, the documents it should use, what a useful contribution looks like, and when it may speak. An idea generator might contribute when invited. A fact checker might request a turn when a decision depends on an unsupported claim. A recorder might briefly confirm a commitment. Choose how the host can pause it.
For a simple action recorder, try this design: it makes a little sound and displays each new to-do with the assigned person so everyone can check it immediately. Have the AI verify which parts your tool supports. The sound-and-display combination may need a custom addition; a short spoken confirmation can serve the same purpose. Keep proposed tasks distinct from commitments people have accepted.
Rehearse with a small group on the actual meeting platform. Test that everyone can hear the AI and see the right output. Correct a task owner, ask it a question answered by its reference documents, and try a question it cannot answer. If it may interrupt, check when it stays quiet and how to stop it. Fix the parts that fail before using it in the meeting.
Introduce the agent's job at the start of the meeting and use the speaking rules you tested. If you chose the human-hosted version, say when you are bringing an AI contribution into the discussion. Before closing, confirm the resulting tasks, decisions, ideas, or research findings. Ask whether its contribution and timing were useful, then refine the brief for next time.
How it goes
What the AI will do
- Finds your meeting context and company tool options, then researches current commercial and open-source ways to add the chosen role.
- Writes the agent's mission, finds its reference material, guides configuration, and prepares an end-to-end rehearsal.
What to have ready
- A meeting or project, and the kind of contribution you would like from the AI.
- Access or setup help only where the AI cannot use your company's tools itself.
Where you come in
- Choose a role and when the AI may speak; confirm any needed tool or integration choice.
- Rehearse, introduce the AI at the meeting, correct its contributions, and judge whether it helped.
What you'll get
A reusable meeting-agent role and a real example of its contribution, with the speaking method and any remaining implementation gaps clearly identified.
You're done when
The AI has made a useful audible contribution in a real meeting—directly or through the human-hosted version you chose—and you have a corrected record and reusable role brief. A silent transcript or after-meeting summary alone does not complete this move.
View full AI instructions
Give AI a speaking role in your meetings.
Give an AI a job in the conversation. It might confirm a new to-do while everyone can correct the owner, keep track of decisions, suggest an idea when the group gets stuck, or check a claim against the documents. A more ambitious agent can arrive with a mission and reference material, then ask to speak when it has something useful to add.
This is a fast-moving area. Some meeting assistants capture notes or show private advice; others can participate by voice. Current examples include Otter's account-dependent voice interaction in Zoom and custom speaking bots built with Attendee. Teams Facilitator provides live notes and chat, while Otter Live Assist provides private coaching cards. Use the prompt below to check what your company can use now and which commercial or open-source route fits your role.
Step 1 — For the user
Choose one job for the AI in an upcoming meeting. Start with the contribution you want: confirming to-dos, recording key decisions, offering ideas, or researching facts. For a specialist role, choose a concrete mission such as checking proposals against the launch brief. The prompt below can investigate your company's tools and guide the setup through the meeting.
Step 2 — For the user
Find a route that can do that job during your actual meeting. A note-taking bot, a private coaching panel, and a participant that speaks to everyone are different experiences.
An available meeting agent
Otter documents voice interaction in Zoom, including action-item commands, but account access must be checked. Start here if your company already uses it and the role fits its supported controls.
https://help.otter.ai/hc/en-us/articles/30839137508631-Otter-AI-Chat-with-voice
Your existing assistant, with a human bringing it into the room
An approved assistant can work from accessible meeting context and give you an idea or sourced answer to read aloud. Teams Facilitator's live notes/chat is one possible starting point. Be clear whether the AI receives a live feed or only what you give it.
A custom speaking participant
Attendee provides an open-source meeting-bot API and a live voice-agent example for Zoom, Teams, and Google Meet. This route needs configuration and development; have the AI prepare the role, build or handoff brief, and test rather than treating the repository as an installed product.
https://github.com/attendee-labs/voice-agent-example
Step 3 — For the user
Give the agent a short brief: its mission, the documents it should use, what a useful contribution looks like, and when it may speak. An idea generator might contribute when invited. A fact checker might request a turn when a decision depends on an unsupported claim. A recorder might briefly confirm a commitment. Choose how the host can pause it.
Step 4 — For the user
For a simple action recorder, try this design: it makes a little sound and displays each new to-do with the assigned person so everyone can check it immediately. Have the AI verify which parts your tool supports. The sound-and-display combination may need a custom addition; a short spoken confirmation can serve the same purpose. Keep proposed tasks distinct from commitments people have accepted.
Step 5 — For the user
Rehearse with a small group on the actual meeting platform. Test that everyone can hear the AI and see the right output. Correct a task owner, ask it a question answered by its reference documents, and try a question it cannot answer. If it may interrupt, check when it stays quiet and how to stop it. Fix the parts that fail before using it in the meeting.
Step 6 — For the user
Introduce the agent's job at the start of the meeting and use the speaking rules you tested. If you chose the human-hosted version, say when you are bringing an AI contribution into the discussion. Before closing, confirm the resulting tasks, decisions, ideas, or research findings. Ask whether its contribution and timing were useful, then refine the brief for next time.
Done when
The AI has made a useful audible contribution in a real meeting—directly or through the human-hosted version you chose—and you have a corrected record and reusable role brief. A silent transcript or after-meeting summary alone does not complete this move.
Guidance for the AI
These are general instructions, not a description of this user's particular AI system, tools, access, or circumstances. Adapt the method to the user's goal and the capabilities you can actually use, while preserving the activity's essential sequence, review points, and intended result.
Start with the activity and the user's context
Read the complete activity and guide the user through it. Use the activity prompt yourself; the user does not need to paste it again. Keep the responsibilities of the user, the AI, and any colleague clear.
Begin with what the user has already supplied. Establish the intended result and resolve missing context through focused questions when needed. Work within the activity's stated scope, permissions, required confirmations, and stopping points, including any deliberate order for introducing information. Make progress with the available information and reasonable, clearly identified assumptions where the activity allows them.
Check the capabilities needed for this work
Inspect relevant available tools, connectors, and data. Use small, relevant, non-destructive checks to establish whether necessary access or a capability actually works. Naming a product does not establish access to it. Let this activity determine what needs checking; do not impose a universal preliminary questionnaire or a full system audit.
When useful access is missing:
- Explain what you cannot reach and how that material would help.
- Investigate practical setup, connection, export, or other access alternatives before asking the user to gather material or troubleshoot.
- If progress depends on access, request only the specific user action or permission needed to enable it or provide the material another way. Continue useful independent work while awaiting the response.
- Offer to walk the user through an export or upload. Welcome files, links, copied text, rough notes, or an interview as supported by your system and the activity, and use the best available route. If the missing access is optional, continue with what is available.
Ask focused questions and respect the answers
Ask questions only when necessary to complete the work well or satisfy the activity's required review points. Build on answers already supplied. Ask one question at a time, or a small group of closely related questions. Wait on work that depends on the answer, and continue useful work that does not. Explain unfamiliar language with examples where helpful.
Confirm answers or decisions you supply on the user's behalf before treating them as the user's final answers or decisions. This does not require separate approval of every sourced finding; follow the activity's specific review requirements.
Keep evidence and independent contributions distinct
Distinguish evidence, inference, and missing information. Check your work against the source material.
Use independent agents when your system actually supports delegation and independent investigation or review would improve the work:
- Give each agent a bounded brief and the context appropriate to its role.
- Keep discovery independent until synthesis where the activity requires it. Preserve any requirement for the user or colleagues to contribute before seeing other explanations.
- Follow the activity's stated fallback and any required confirmation when independent agents or fresh contexts are unavailable. Preserve reviewer isolation where required; describe shared-context or sequential passes accurately.
Produce, check, and revise the result
Do the substantive investigation and production. Carry the activity through to the most complete, usable result within its scope, rather than ending with a plan or handing the main work back to the user. When missing information does not prevent useful work, proceed with clearly identified assumptions where the activity permits them.
Check the result against the activity's completion criteria. At each substantive delivery, state what is complete and any material unfinished work, assumptions, helpful missing information, and remaining uncertainty. Explain how these affect what the user can use, and help the user revise the result.
Preserve the activity's stopping point. Do not turn a rehearsal, investigation, or test-design activity into execution that the user has not authorized.
Your AI prompt
Help me give an AI a useful, audible role in a real meeting. This is a fast-moving area: first find what is available in my company and what current commercial or open-source tools could make it work. Then help me configure one bounded role, test it, and use it in a meeting. Do not stop at a list of tools.
Find the meeting context and our tools
Use the meeting link, project, or rough description I have supplied. Search accessible calendar entries, agendas, relevant email and chat threads, earlier meeting transcripts, and project documents to establish the meeting's purpose, participants, outstanding decisions, existing commitments, and useful reference material. Look for participants' actual questions and concerns. Keep documented positions separate from inferred ones.
Search accessible company software catalogs, IT guidance, license records, approved-app lists, and prior support messages for the meeting platform, AI assistants, bot/app permissions, and existing research or document connectors. Actually inspect available access. If you cannot inspect an account or connect to the meeting, say exactly what is missing and investigate the setup path before asking me to do anything.
Ask one focused question at a time for missing choices: the meeting, the job the AI should perform, whether it needs to speak directly, and who should hear or see its contributions. Use context already supplied; do not make me prepare a document packet first.
Check the current product capabilities
Search current primary documentation, official app listings, and source repositories. Date the findings and distinguish shipping functionality, account-gated rollout, preview, demonstration, and custom development.
Check our existing platform's options first, then compare the relevant commercial products and at least one credible open-source route if a custom role would benefit us. Starting references to recheck:
- Teams Facilitator: https://support.microsoft.com/en-us/teams/platform/frequently-asked-questions-about-agents-in-microsoft-teams
- Otter voice interaction in Zoom: https://help.otter.ai/hc/en-us/articles/30839137508631-Otter-AI-Chat-with-voice
- Otter voice actions: https://help.otter.ai/hc/en-us/articles/30910450353431-Otter-AI-Chat-voice-commands-and-actions
- Otter Live Assist and Agent Builder: https://otter.ai/blog/otter-ai-introduces-live-assist-the-first-live-coaching-agent-for-every-call
- Attendee source: https://github.com/attendee-labs/attendee
- Attendee speaking-agent example: https://github.com/attendee-labs/voice-agent-example
- Attendee media capabilities: https://attendee.dev/changelog
- Fireflies Voice Agents: https://guide.fireflies.ai/articles/6333219586-how-to-view-your-voice-agent-meetings
Verify the functions independently:
- Hear the current meeting through live audio or a live transcript.
- Generate responses while the meeting is happening.
- Speak audibly to everyone, answer voice commands, or only display private/chat responses.
- Read the reference documents and search sources needed for this specific role.
- Display an action item with its owner, and support corrections while the group is present.
- Wait for a cue, raise a hand/request a turn, or interrupt under specified conditions.
- Join this exact platform and meeting type using our account and device setup.
A product that transcribes meetings or hosts its own AI interview does not necessarily join our meeting as a speaking collaborator. An MCP connection to past transcripts does not establish live listening. A generic voice model does not establish meeting integration. Verify the whole path.
Identify licenses, rollout status, who must install/enable/admit the app, where its audio and visible output appear, current cost when established, and the work required to create a custom integration. Do not assume open source means no hosting, speech-model, or operating costs.
Give me a short comparison and recommend a route for my chosen role. If no direct speaking route is available, propose a clearly labeled human-hosted version: an approved AI receives selected context or available live transcript, and a person reads its contribution aloud or plays approved audio. Explain what that version cannot monitor or do on its own. Wait for my choice if the fallback changes the intended activity.
Create one useful role
Begin with one role; offer more ambitious behavior only if it serves the meeting. Write a ready-to-use agent brief containing the mission, reference material with dates/source links, available tools, expected output, speaking rules, and completion point. Configure those settings where the product permits it. Where it does not accept custom instructions, turn the brief into the exact supported settings and commands; do not imply pasting a prompt changes a fixed product's behavior.
Possible roles:
- Action recorder: capture a proposed task, named owner, due date only if supplied, and a short supporting quote/time. Show it while people can correct it. Ask for owner/date clarification only when needed; do not invent assignments.
- Decision recorder: separate an agreed decision from a suggestion or unresolved debate. Record who confirmed it, its scope, and the relevant source/time; bring back ambiguous items for the group to settle.
- Idea generator: use the actual meeting question and constraints to offer a few distinct ideas, explain what is new about each, and mark them as proposals. Contribute on invitation or at an agreed agenda point.
- Fact checker/researcher: search the supplied company record and approved external primary sources for claims that matter to the decision. Show the exact claim, what the sources support or contradict, their dates, and unresolved uncertainty. Separate checking against supplied material from doing a live external search.
- A specialist with a mission: for example, compare proposals against an agreed launch brief or customer requirements. Ground the mission in actual documents and define the conditions that justify seeking a turn.
Preserve this possible simple design: when the AI captures a to-do, it makes a little sound and displays the task and assignee so the group can confirm or correct them. Check whether the selected tool actually does that. If it supports the task display but not the sound, state that precisely; offer an audible spoken confirmation or a small custom addition only if I want it. Do not attribute an unverified chime feature to a vendor.
Set a speaking policy with me. It can be invitation-only, at named agenda points, or allowed to request a turn when a specific trigger occurs. For an interrupting agent, define those triggers, how it waits for a natural pause, how brief it should be, and how the host can silence it. Do not convert permission to speak into permission to assign people work, send messages, or change company systems.
Set up the chosen route
Prepare the role brief and load the relevant reference material within the allowed tool. Verify that the agent can actually retrieve a named fact from the material and identify its source.
For an available commercial agent, do permitted configuration yourself, then give the host only the required enablement, joining, and audio actions you cannot do. Prepare a short introduction so participants know the agent's role and how to correct or pause it.
For a custom route, produce an implementation brief for the existing technical owner or coding assistant: live meeting input, reference retrieval, model/tool calls, turn-taking, speech output, visible task/decision output, correction storage, and a host stop control. Specify which parts already exist in the selected project and which must be built. If implementation is authorized and you have the capability, build the smallest version and verify it. Otherwise deliver the exact bounded handoff and its acceptance test; do not call a repository link or untested demo a working agent.
For the human-hosted version, prepare short prompts and an easy way for the host to provide the selected context and bring the result into the room. Keep the substantive research and drafting with the AI. Do not require a person to secretly transcribe the whole meeting or pretend the AI heard what was never supplied.
Ask for any necessary install, purchase, external connection, or actual bot-invitation authorization before taking that action. Prepare the concrete configuration first. Keep access requests distinct from the finished meeting outcome.
Rehearse the role end to end
Prepare a short realistic rehearsal using the same accounts, platform, and output channel as the real meeting. Confirm everyone intended to hear the AI can hear it, and that visible output reaches the intended people. Check that the system does not listen to its own output and repeatedly respond.
Exercise the chosen role:
- For tasks/decisions, include one genuine commitment, one hypothetical suggestion, a corrected assignee, and an unspecified deadline. Check the displayed record after the correction.
- For ideas, pose an actual decision question, then add a constraint and check whether its suggestion changes.
- For research, use one supported claim, one incorrect claim, and one claim the provided sources cannot settle. Check citations, source dates, and its willingness to leave the last one unresolved.
- If proactive speech is enabled, demonstrate an allowed trigger, ordinary discussion where it should stay quiet, a person speaking over it, and the host's pause/stop command.
Observe results directly when possible; otherwise ask for the specific missing evidence. Diagnose failures and repeat the affected test after a fix. Do not claim it is ready because configuration saved successfully. If the route cannot perform its speaking role, agree a reduced role or fallback before the real meeting.
Use it in a meeting and evaluate the contribution
Support the actual meeting only through capabilities you have verified. Make clear whether you are connected live or helping before/after. At the end, collect the confirmed tasks or decisions, proposed ideas, or sourced research results appropriate to the role.
Ask the group whether the AI contributed something useful and whether its timing helped. Record one or two concrete examples, corrections, and the change to make before the next meeting. Leave an updated reusable role brief and setup instructions. Mark completion only after a real meeting includes the chosen audible AI contribution, either directly or through the explicitly agreed human-hosted version. An implementation brief, silent recorder, or post-meeting summary alone is a partial result.
These instructions were written in September 2026 using the frontier models available at the time. You may now have better tools or models, or restrictions that make some tasks harder. Your AI should adapt to the tools and access you actually have and find the best available way to help. Everything here is a starting suggestion for you and your AI, not a proven best practice for every setup. Feel free to rewrite any of it to better meet your needs.
These instructions were written in September 2026 using the frontier models available at the time. You may now have better tools or models, or restrictions that make some tasks harder. Your AI should adapt to the tools and access you actually have and find the best available way to help. Everything here is a starting suggestion for you and your AI, not a proven best practice for every setup. Feel free to rewrite any of it to better meet your needs.