A new lead fills in your quote form at 4:40 on a Friday. Before anyone calls them back, somebody has to look up the business, check whether they are already in the CRM, write a note, set a task and decide how urgent it is. None of that is hard. It is also the work that gets skipped when the afternoon is busy.
That kind of job suits a Managed Agent. You describe the job in plain English in Agent Studio, choose what the agent may read and change, and pick what starts it. This guide lists 30 jobs worth trying, grouped by team. Each one names the trigger and gives a starter prompt. Most also say what has to be in place first and where a person should stay involved.
Treat every prompt as a draft to test in your own account. They are starting points written against the documented triggers and skills, not recipes we have run end to end, and your forms, pipelines, tags and connected apps will differ. Replace anything in [BRACKETS] with your own names and numbers, run it in the test area, and read the result before you publish the agent. In the prompts, [OWNER] is the person who should get a task, [CONNECTOR] is the connected app that does the searching, and [DESTINATION] is where a report is saved, such as a connected Google Doc.
Before you build one
A workflow follows the same steps every time. An agent reads the situation and decides what to write or which step applies, inside the permissions you give it. If the job is “when this happens, always do that”, a workflow is simpler and more predictable. Use an agent when the job needs reading, judging or writing.
What can start an agent
| Trigger | Starts when | Useful detail |
|---|---|---|
| Chat | You message the agent | Best for research and drafting on request |
| Form Submitted | A chosen form is submitted | One run per submission |
| Survey Submitted | A chosen survey is submitted | Confirm it appears in your trigger picker |
| Tag Changed | A tag is added or removed | Lets staff start an agent by tagging a contact |
| Contact Created | A contact is added | No filters |
| Opportunity Created | A deal is created | Filter by pipeline and stage |
| Opportunity Status Changed | A deal changes status | Open, Won, Lost or Abandoned |
| Customer Booked Appointment | A booking lands on a chosen calendar | Pick at least one calendar |
| Appointment Status Changed | An appointment changes status | Confirmed, Cancelled, Showed, No Show or Rescheduled |
| Schedule | A time arrives | Once, on an interval or by cron expression, in the account time zone |
| Invoked from a workflow | A workflow reaches the Invoke Managed Agent action | Passes the contact and workflow data |
| Inbound webhook | Another system calls the agent | Verified with a shared token |
An agent can also pause and resume. Scheduled follow-ups let it wait for a set delay, or until a set time before or after an appointment, then pick up with the earlier context. Each wait can run from 5 minutes to 30 days.
What a run costs
One run is one chat session, one event invocation or one scheduled invocation. The test area includes a monthly allowance of free runs. In production, each AI Employee tier includes a monthly allowance of Managed Agent runs. Without AI Employee, runs are pay-per-use and draw on your plan’s monthly usage credit first. The AI billing section has the full terms.
Count runs before you choose a trigger. A weekday digest uses about 22 runs a month. An agent on your main lead form uses one run per submission, so 150 submissions is 150 runs. Compare that with your monthly allowance. A trigger with no filter, such as Contact Created, starts a run for every new contact, so prefer one you can narrow to a single form, pipeline or calendar. What happens once an allowance is used depends on your account’s usage-limit settings. Check them before you publish an event-triggered agent.
Rules that apply to all 30
- Approval is something you write. Outside Computer, there is no general approval step for CRM actions, so the prompt has to say what the agent may do alone and what it must leave as a draft or a task for a person. Most prompts below stop at a draft.
- Messages go through workflows you have already approved. In these examples the agent adds tags and notes and starts a named workflow. It does not compose and send customer messages on its own.
- Tests can take real actions. A test run can update records and start workflows, and it skips waits. Test with your own contact record.
- Permissions are the real limit. A prompt line such as “do not change any record” is an instruction. The agent’s permissions are the control. Give each agent only the CRM access its job needs, and leave web search and image generation off unless the job uses them.
- Watch the first runs. After you publish an agent, open Activity and Agent Logs to see what each run did.
- Computer works in chat and manual runs only. Browser, Files and Terminal are not available to event, workflow or scheduled runs. Triggered research needs a connector such as Firecrawl instead. See how Computer works.
- Apps have to be connected first. Prompts that mention Google Sheets, Google Docs, Apollo, Firecrawl, Asana, ClickUp, Monday.com or Fireflies assume that app is connected to the agent with the tools it needs. Check the Apps list in your account first.
Put these lines at the end of every prompt, or save them once as a custom skill if your account shows that option:
House rules:
- Use only information in the CRM, the attached knowledge base and the tools you were given. If something cannot be verified, write "unknown".
- Never invent prices, dates, availability, discounts or policies.
- Do not contact, or start a workflow that sends messages to, anyone marked Do Not Disturb or unsubscribed.
- Do not delete records, change deal values or owners, or take payments.
- If you are unsure, stop and say what you need. Create a task for [OWNER] only if this prompt lets you create tasks.Sales and lead management
1. Prospect finder
A commercial cleaning company wants a list of property managers within 20 miles. The agent searches a prospecting connector, removes duplicates and anyone already in the CRM, and saves the list for review. Start it from chat. It needs Apollo or a similar connector and a Google Sheet. Nobody on the list gets contacted by the agent.
You find prospects for [BUSINESS]. When I ask, search [CONNECTOR] for businesses matching: industry [INDUSTRY], location [AREA], size [SIZE]. Return up to [NUMBER] per request. For each, record name, website, phone, contact name and role, and the source. Write "unknown" in any field that was not found. Remove duplicates, and remove any business that already exists in the CRM. Add the rest to the sheet [SHEET NAME]. Do not create contacts and do not send any outreach.2. Scoring new leads
When a quote form comes in, the agent looks up the business and scores the lead against your criteria, so the owner sees the best ones first. Trigger: Form Submitted. It needs written criteria with weights and a research connector. If you have not defined criteria yet, Lead Scoring 101 covers how to choose them.
Run when [FORM NAME] is submitted. Review the answers and the contact record. Research the business with [CONNECTOR]. Score the lead from 0 to 100 using these criteria and weights: [CRITERIA]. Base every point on evidence you can cite. Add a note with the score, a High, Medium or Low label, the evidence and one suggested next step. Add the tag that matches the label: [HIGH TAG], [MEDIUM TAG] or [LOW TAG]. If the label is High, create a task for [OWNER] due today.3. New opportunity brief
A rep opens a new opportunity and gets a short brief in the notes: who the customer is, what they asked for, what has been said so far and what is missing. Trigger: Opportunity Created, filtered to one pipeline. The agent writes notes and tasks and changes nothing else.
Run when an opportunity is created in [PIPELINE]. Read the contact, the opportunity and the conversation history. Write a brief as a note with these parts: who they are, what they asked for, known constraints, history with us, open questions. Put what the record shows under one heading and your own ideas under another. If the opportunity has no follow-up task, add one for the opportunity's owner. Leave the stage, value and owner as they are.4. Service-matched welcome
A landscaping company gets inquiries for design, maintenance and irrigation. The agent works out which service the lead wants, records it and starts the matching welcome workflow. Trigger: Form Submitted. It needs a welcome workflow for each service, a custom field for the service and one new-lead tag. Each welcome workflow should use only the channels the lead agreed to on the form.
Run when [FORM NAME] is submitted. If the contact already has the tag [NEW LEAD TAG] from an earlier submission, add a note recording the new request and stop. Otherwise decide which service the lead is asking about: [SERVICE LIST]. If it is unclear, use "General". Set the field [FIELD] to that service, add the tag [NEW LEAD TAG] and start the workflow that matches the service: [SERVICE TO WORKFLOW MAP].5. Appointment prep
A bookkeeper who meets new clients back to back on Tuesdays gets a short brief before each meeting. Trigger: Customer Booked Appointment on a chosen calendar, with a scheduled follow-up set for a fixed time before the meeting. The agent should check the appointment again when it resumes, because bookings move.
Run when an appointment is booked on [CALENDAR]. Wait until [TIME] before the appointment. Then check the appointment again. Stop if the booking was cancelled. If the time moved, work from the new one. Read the contact, notes, conversations and any open opportunity. Add a note titled "Call prep" with: reason for the call, background, what they have told us, open issues and five questions to ask. Do not message the contact.6. Missed-appointment rebooking
Someone misses a consultation. The agent starts your rebooking workflow, tags the contact and leaves a task if the booking was a valuable one. Trigger: Appointment Status Changed, filtered to No Show. It needs a rebooking workflow with a neutral message that does not guess why the person missed it.
Run when an appointment on [CALENDAR] changes to No Show. If the contact already has a future appointment or has opted out, stop. If the contact already has the tag [NO SHOW TAG], stop and create a task for [OWNER] instead. Your rebooking workflow removes that tag after [DAYS] days. Otherwise add the tag [NO SHOW TAG] and start the workflow [REBOOKING WORKFLOW]. If the contact has an open opportunity above [AMOUNT], also create a call task for the opportunity's owner.7. Lead follow-up sequence
A tutoring center offers trial lessons, and many parents who said they would think it over never come back. A staff member adds a tag, and the agent starts up to three check-in workflows with waits in between, stopping when the lead replies, books or opts out. Trigger: Tag Changed. It needs one approved workflow per touch and scheduled follow-ups. A test run skips the waits, so all three workflows start one after another on your test contact.
Run when the tag [FOLLOW UP TAG] is added. Before each step, stop if the contact has replied, booked, become a customer or opted out, or if the tag was removed. Step 1: start [WORKFLOW 1]. Wait [DAYS] days. Step 2: start [WORKFLOW 2]. Wait [DAYS] days. Step 3: start [WORKFLOW 3]. After step 3, add a note saying all three steps were started.8. Inactive lead re-engagement
Once a week the agent builds a short list of leads with no activity for 90 days and explains why each one qualified. A person approves the list before any campaign goes out. Trigger: Schedule. Finding people by inactivity depends on what the agent can search, so compare its first list with a smart list you trust. Read the database reactivation guide before you contact anyone on it, because eligibility and consent come first.
Run every [DAY] at [TIME]. Find up to [NUMBER] contacts whose [LAST ACTIVITY FIELD] shows no activity for [DAYS] days. If you cannot see an activity date, say so and stop. Skip current customers, anyone mid-conversation, unsubscribed or Do Not Disturb contacts, and contacts already carrying [CAMPAIGN TAG] or [REVIEW TAG]. Group them by [INTEREST FIELD]. Add the tag [REVIEW TAG] and a note giving the reason each was selected. Create one task for [OWNER] to review the list. Do not start any workflow.9. Lost deal review
When a deal is marked Lost, the agent reads the thread and records a reason only if the conversation supports one. If the reason is timing, it sets a task to revisit the deal. Trigger: Opportunity Status Changed, filtered to Lost.
Run when an opportunity in [PIPELINE] changes to Lost. Read the notes and conversations. Choose one reason: price, timing, chose another provider or scope. If the record does not clearly support one, use "unknown". Add a note with the reason and, unless it is unknown, the line from the record that backs it up. When the reason is timing and the contact has no open complaint or opt-out, set a follow-up task for the opportunity's owner [DAYS] days out.10. Daily pipeline digest
Every weekday morning the owner gets one page: open value, the biggest deals, and deals with no recorded activity in a week. Trigger: Schedule. It needs somewhere to save the report, such as a connected Google Doc. Give it read access to opportunities and nothing more. That makes it a good first agent, because the job changes nothing in the CRM.
Run every weekday at [TIME]. Read all open opportunities in [PIPELINE]. Report: total open value, the [NUMBER] largest deals, deals with no recorded activity for [DAYS] days, and deals with no follow-up task. Flag missing values or owners. An empty activity log means no activity was recorded, so say that and do not conclude that nobody made contact. Save the report to [DESTINATION]. Do not change any record.11. CRM data check
Each month the agent reviews one segment and lists records with missing fields, mismatched tags and likely duplicates. It fixes capitalization in the fields you name and leaves everything else for review. Trigger: Schedule. Pair it with the year-end cleanup checklist, which explains why merges and deletions stay with a person.
Run on the first [DAY] of each month. Review contacts in [SEGMENT]. List records missing [REQUIRED FIELDS], records whose tags conflict with these rules: [TAG RULES], and pairs that look like duplicates, with the reason. You may fix capitalization in [FIELDS] and nothing else. Never merge or delete. Save the list to [SHEET NAME] and create a review task for [OWNER].Marketing and content
In the content jobs here (12 to 15), the agent drafts and a person publishes. Scheduling posts and sending campaigns stay manual, so every piece gets read first. The rest of the group covers reports, review requests and survey feedback.
12. Weekly social drafts
A bakery wants five posts a week in its own voice. The agent drafts captions and image briefs from a knowledge base that holds the brand guidelines and content themes, and avoids repeating recent posts. Trigger: Schedule. Review the drafts, then schedule them in the social planner.
Run every [DAY]. Using the knowledge base [BRAND KB], draft [NUMBER] posts for [PLATFORMS] for next week, covering these themes: [THEMES]. For each post give the caption, up to [HASHTAG LIMIT] hashtags, one call to action and a one-sentence image brief. Read the earlier entries in [DOC NAME] and do not repeat a topic from the last four weeks. Add the drafts to [DOC NAME] under next week's date. Do not publish or schedule anything.13. Email campaign draft
You tell the agent the audience and the goal, and it returns a campaign ready to edit: two subject lines, preview text, body and one follow-up. Trigger: Chat. A person builds and sends it in email marketing.
When I describe a campaign, ask me for the audience, the goal and the offer if I left any out. Then draft: a one-line objective, a plain description of who should receive it, two subject lines, preview text, the body, one call to action and one follow-up email for people who did not open. Use the knowledge base [BRAND KB] for tone and facts. Save the draft to [DOC NAME]. Do not create or send a campaign.14. Weekly creative concepts
Each week the agent produces a few ad concepts for the current promotion, with headlines and images. Trigger: Schedule. Image generation is a separate switch on the agent and stays off until you turn it on. Nothing is published.
Run every [DAY]. Create [NUMBER] creative concepts for [OFFER] aimed at [AUDIENCE]. For each concept write a headline, a caption and a suggested placement, and generate one image that follows [BRAND KB], aiming for [DIMENSIONS] if the image tool supports sizes. Save the images to Media Storage in the folder [FOLDER]. Finish with a summary listing each concept and its file name. Do not publish anything.15. Content repurposing
A dental practice records a talk on gum care for patients, and its approved transcript becomes three LinkedIn posts, a newsletter draft and two short video scripts. Trigger: Chat, with the source pasted in or attached. The agent should work only from the source so it adds no new claims.
When I give you a source, use only that source. Write: [POST COUNT] LinkedIn posts, one newsletter draft and [SCRIPT COUNT] scripts for videos under [SECONDS] seconds. Adapt length and tone to each channel. Do not add facts, numbers or quotes that are not in the source. Save everything to [DOC NAME] with a heading for each format.16. Weekly competitor check
Each Monday the agent reads the public pricing and service pages of three competitors and reports what changed since last week, with links. Trigger: Schedule. Because scheduled runs cannot use Computer, this one needs Firecrawl or another crawling connector.
Run every [DAY]. Use [CONNECTOR] to read these public pages: [URL LIST]. Record the services, prices and announcements shown. Compare them with last week's entry in [SHEET NAME]. If there is no earlier entry, record this week as the baseline and report no changes. Report each change with the page link, and mark it as either stated on the page or your own inference. Add this week's entry to the sheet. Do not log in to any site.17. Ad performance report
The agent pulls Facebook Ads and Google Ads reporting for the week, compares it with the week before and lists the strongest and weakest campaigns. Trigger: Schedule. It needs connected ad accounts, ad reporting permission for the agent and somewhere to save the report.
Run every [DAY]. Pull Facebook Ads and Google Ads reporting for the last [PERIOD] and for the period before it. Report spend, leads and cost per lead for each campaign, using only the fields the reporting returns. Name the top and bottom [NUMBER] campaigns by [METRIC]. Suggest up to three changes and give the number behind each one. Do not predict results. Save the report to [DESTINATION]. Do not change any campaign or budget.18. Lead source report
Monthly, the agent counts leads, qualified leads and won deals by source, so you can see which sources your customers came from. Trigger: Schedule. The numbers are only as good as your source tracking, so missing sources are reported as Unknown.
Run on the first [DAY] of each month. For last month, group contacts and opportunities by lead source. For each source report the number of leads, the number that reached [QUALIFIED STAGE] and the number won. Put records with no source under "Unknown" and report that count. If you could not read every record, say the counts are partial. Describe patterns you see, and do not claim a cause. Save the report to [DESTINATION]. Do not change any record.19. Review request check
After a visit is marked Showed, the agent checks that the customer has not been asked recently and has not opted out, then starts your review workflow. Trigger: Appointment Status Changed, filtered to Showed. Every eligible customer gets asked. Do not have it pick only the happy ones, and do not offer anything in return for a rating. The workflow itself lives in reputation management.
Run when an appointment on [CALENDAR] changes to Showed. Stop if the contact has the tag [REVIEW REQUESTED TAG] or has opted out. Your review workflow removes that tag after [DAYS] days. Otherwise start the workflow [REVIEW WORKFLOW], add the tag [REVIEW REQUESTED TAG] and add a note that the request was started. Apply the same rule to every customer. Do not choose customers by how satisfied they seem, and do not offer anything in exchange for a review.20. Survey response triage
When a satisfaction survey arrives, the agent summarizes it, classifies it and creates a task for anything that reads as a complaint. Trigger: Survey Submitted. If your trigger picker does not list it, have a workflow invoke the agent when the survey comes in.
Run when [SURVEY NAME] is submitted. Summarize the response in three sentences. Classify it as one of: [CATEGORIES]. Label the tone positive, neutral or negative. Add the summary as a note and add the tag that matches the category: [CATEGORY TO TAG MAP]. If it contains a complaint, create a task for [OWNER] due within [HOURS] hours. Do not reply to the customer or commit to any refund, credit or special treatment.Support, onboarding and retention
21. Customer onboarding coordinator
A deal is won and the same six setup tasks need creating every time, plus a summary for whoever takes over. A workflow could create the tasks on its own. The agent is here for the handover note, which takes reading the record. Trigger: Opportunity Status Changed, filtered to Won.
Run when an opportunity in [PIPELINE] changes to Won. If the contact already has the tag [ONBOARDING TAG], stop. Otherwise add the tags [CUSTOMER TAG] and [ONBOARDING TAG]. Write a handover note covering what was sold, what was promised, key dates stated in the record, and any information still missing. Create these tasks for [OWNER]: [TASK LIST]. Start the workflow [ONBOARDING WORKFLOW].22. Knowledge base assistant for staff
A new team member asks how refunds work, and the agent answers from your documented policies and names the document it used. Trigger: Chat. It needs a knowledge base that is current, and staff should open the named document before acting on an answer about refunds or money. For answering customers directly on chat or SMS, Conversation AI is the product built for that.
Answer staff questions using only the knowledge base [SUPPORT KB]. Name the document each answer came from when the search result shows it. If the question is unclear, ask one clarifying question. If the knowledge base does not cover it, say so, do not guess, and create a task for [OWNER] titled "Knowledge base gap" with the question.23. Support triage
Support requests arrive through a form. The agent classifies each one, rates urgency against your written criteria and creates a routed task. Trigger: Form Submitted. Security problems, billing disputes and outages go straight to a person.
Run when [SUPPORT FORM] is submitted. Classify the request as one of: [CATEGORIES]. Rate urgency using these rules: [SEVERITY RULES]. Write a two-sentence summary. Add the tag that matches the category: [CATEGORY TO TAG MAP]. If no open task exists for the same issue, create one and assign it using: [ROUTING RULES]. Anything about account security, a billing dispute or an outage is always urgent and goes to [OWNER]. Leave replies and closing the request to a person.24. Community moderation helper
A moderator pastes in a post and asks whether it breaks the rules. The agent compares it with your guidelines and recommends an action. Trigger: Chat. It advises, and a person removes posts or members. We have kept this manual on purpose, because the trigger list above has no event for a new community post.
I will paste a community post or comment. Compare it with the guidelines in [GUIDELINES KB]. Tell me whether it breaks a rule, which rule, and what you recommend: leave it, edit, remove or escalate. If it is a borderline case, say that and explain both readings. Do not take any action yourself.25. Course progress follow-up
Each week the agent finds learners who have not reached a milestone and starts a reminder workflow for them. Trigger: Schedule. It uses the Courses skill, which can read enrollment and progress. Check what it returns in your account. If you only see overall progress, base the rule on that and not on a milestone.
Run every [DAY]. In [COURSE], find learners who have not completed [MILESTONE]. If you cannot see milestones, use overall progress below [PERCENT]. Skip anyone who has opted out and anyone who currently has the tag [REMINDER TAG]. The learner workflow removes that tag when it finishes. For the rest, add the tag [REMINDER TAG] and start the workflow [LEARNER WORKFLOW]. Add a note with their current progress. Do not change any enrollment.26. Renewal prep
A workflow tags a customer 60 days before renewal. The agent prepares a brief and a draft message for the contact’s owner to edit. Trigger: Tag Changed. It needs renewal dates stored in the CRM. Leave subscription and payment access off for this agent.
Run when the tag [RENEWAL TAG] is added. Read the contact, notes, conversations and opportunity. Write a note titled "Renewal brief" with: current agreement details that appear in the record, concerns raised this year, open issues and questions to ask. Then draft a renewal message and add it to the same note, marked DRAFT. Create a task for the contact's owner. Do not send the message. Do not state terms, prices or dates that are not in the record.Operations and team productivity
27. Hiring coordinator
Applications come through a form. The agent summarizes each one against the written requirements for the role and logs it in a tracker. Trigger: Form Submitted. People make every decision, and the recruiter reads the full application, not only the summary. The agent does not rank or reject anyone. Some places regulate automated tools in hiring, so check the rules where you hire.
Run when [APPLICATION FORM] is submitted. Compare the answers with these role requirements: [REQUIREMENTS]. For each requirement, quote the answer that addresses it, or write "not shown". Do not judge whether the applicant is suitable. Use only job-related information, and draw no conclusion from a name, address, photo or gap in work history. Do not infer age, gender, ethnicity, health, family status or any similar characteristic. Do not rank, score or reject anyone. Add a row to [TRACKER] and create a task for [RECRUITER].28. Meeting notes to tasks
After a client call, the agent turns the transcript into decisions, open questions and tasks in your project tool. Trigger: Chat. The prompt is written for chat, so reword the opening if a workflow starts it when your notes arrive. It needs a transcript, from Fireflies or pasted in, and a connected task tool.
When I give you a transcript or name a Fireflies meeting, extract: the purpose of the meeting, decisions made, open questions and action items. For each action item record the owner and due date only if they were stated. Mark the others "owner needed" or "date needed". Create the tasks in [TASK TOOL] under [PROJECT], skipping any you can see already exist. Reply in this chat with the summary and links to the tasks.29. Project kickoff
A project deal is won and delivery needs a checklist, a project and a kickoff agenda. Trigger: Opportunity Status Changed, filtered to Won on your projects pipeline. It needs the scope written down in the opportunity and a project template.
Run when an opportunity in [PROJECT PIPELINE] changes to Won. Read the scope in the opportunity and the standards in [PROJECT KB]. If a project for this client already exists in [TASK TOOL], stop and create a task for [OWNER]. Otherwise create a project from the template [TEMPLATE] and name it after the client. If the tool cannot create a project from a template, create the kickoff tasks one by one and say that you did. Add a kickoff checklist. Draft a kickoff agenda as a note. List anything the scope leaves unclear. Use only the deliverables, dates and prices the scope states.30. SOP and FAQ drafts
A manager hands the agent the notes on how invoices are issued and asks for a procedure. The agent writes it from those notes and flags gaps and contradictions. Trigger: Chat. The draft goes to a Google Doc for review, and your knowledge base stays unchanged until someone approves it.
When I ask for a procedure or an FAQ, use only the sources I name. For a procedure, open with the goal, then give requirements, who is responsible, numbered steps, how to tell it worked and what to do when something goes wrong. For an FAQ write short questions and answers. End with a list of missing information and places where the sources disagree. Save the draft to [DOC NAME]. Leave publishing and the knowledge base to a person.Where to start
Pick one agent that only reads. The daily pipeline digest (10) and the lead source report (18) change nothing in the CRM, so the likely cost of a mistake is a bad report. Run it in the test area first, then compare its reports with what you see in the CRM for a week.
Next, try one that writes notes and tasks but sends nothing, such as the new opportunity brief (3) or support triage (23). Read its runs in Activity and Agent Logs for the first few days. Add an agent that starts a customer-facing workflow, like missed-appointment rebooking (6), only after the workflow itself has been tested with your own contact.
Triggers, limits and billing checked against product documentation on November 13, 2026. Features and names can change, so confirm them in your account before you build.
See what a Managed Agent could take off your list
Bring one of these jobs to a demo and we will walk through the trigger, permissions and review steps it would need.
Explore this in SMBcrm
Give an agent a defined task.
Explore Managed Agents, connected tools, memory, and review controls. Check which capabilities work in manual chats and automated runs.
Explore Managed Agents
