Use cases
AI agent use cases: what teams put agents to work on
The most common AI agent use cases are customer support, email triage, keeping a CRM up to date, scheduled reports, answering questions about your data, and giving a team one private chat for AI models and company tools. Each one pairs a model with a few tools and a rule for which actions need a person’s approval.
Updated · 5 min read
What are the most common AI agent use cases?
These are the jobs teams most often hand to an agent first. Each one reads from and writes to tools you already use, and each one starts from a clear moment: a message, a schedule or a request from another system.
| Use case | What the agent does | What starts it | Typical tools |
|---|---|---|---|
| Website chat | Answers visitors, looks up orders, opens tickets | A visitor’s message | Shopify, Zendesk, HubSpot |
| Support tickets | Reads a new ticket, finds similar ones, drafts the reply | A webhook from the help desk | Zendesk, Freshdesk, Intercom |
| Inbox triage | Sorts and labels mail, summarizes it, drafts replies | A schedule | Gmail, Outlook, Slack |
| CRM upkeep | Creates and updates contacts and deals from conversations | A chat, a form or a workflow | HubSpot, Salesforce, Pipedrive |
| Scheduled reports | Pulls numbers, explains what changed, posts the summary | A schedule | Google Analytics, Google Sheets, Slack |
| Questions about your data | Turns a question into a read-only database query | A chat or a workflow | PostgreSQL, MySQL, Snowflake |
| Team chat | One private chat for every model, your agents and your tools | A person in the chat app | Any connected tool |
| Agents working together | Hands parts of a task to other agents, yours or outside ones | Another agent | Sub-agents, A2A agents |
Website chat
- What the agent does
- Answers visitors, looks up orders, opens tickets
- What starts it
- A visitor’s message
- Typical tools
- Shopify, Zendesk, HubSpot
Support tickets
- What the agent does
- Reads a new ticket, finds similar ones, drafts the reply
- What starts it
- A webhook from the help desk
- Typical tools
- Zendesk, Freshdesk, Intercom
Inbox triage
- What the agent does
- Sorts and labels mail, summarizes it, drafts replies
- What starts it
- A schedule
- Typical tools
- Gmail, Outlook, Slack
CRM upkeep
- What the agent does
- Creates and updates contacts and deals from conversations
- What starts it
- A chat, a form or a workflow
- Typical tools
- HubSpot, Salesforce, Pipedrive
Scheduled reports
- What the agent does
- Pulls numbers, explains what changed, posts the summary
- What starts it
- A schedule
- Typical tools
- Google Analytics, Google Sheets, Slack
Questions about your data
- What the agent does
- Turns a question into a read-only database query
- What starts it
- A chat or a workflow
- Typical tools
- PostgreSQL, MySQL, Snowflake
Team chat
- What the agent does
- One private chat for every model, your agents and your tools
- What starts it
- A person in the chat app
- Typical tools
- Any connected tool
Agents working together
- What the agent does
- Hands parts of a task to other agents, yours or outside ones
- What starts it
- Another agent
- Typical tools
- Sub-agents, A2A agents
The step-by-step guides: customer support, an email assistant, scheduled reports, questions about your database, a private chat for your team, your website and your CRM.
How do you pick your first AI agent use case?
Pick work that happens often, follows rules you could write down, and is easy to check. Good first projects share four traits:
- It repeats. Tens or hundreds of times a week, so the time saved is easy to see.
- The rules fit on a page. If you can explain the job to a new hire in writing, you can explain it to an agent.
- It starts by reading. An agent that only looks things up and drafts can do no harm while you learn how it behaves.
- A person can check the result. A draft reply, a label or a summary is quick to review before anything leaves the building.
A common path is to start with reads and drafts, watch a few days of runs, then add the actions that change things, with approval switched on for each one.
Does every use case need an AI agent?
No. When the steps are always the same, a workflow is cheaper and more predictable: fetch the numbers, ask a model to summarize them, post the result. An agent earns its place when the next step depends on what it finds, such as which tickets to look up before answering.
In MyDeepAgents a workflow can call a tool, make a web request, ask a model, branch, wait, or run an agent for the part that needs judgment. See AI agents vs workflows.
How does an agent get started on its work?
- A person asks. In the website chat, in the team chat app, or in the agent’s test chat.
- A schedule. A workflow runs every hour, every weekday at 9:00 or any cron schedule, in your time zone.
- Another system. Any app that can call a web address starts a workflow through its webhook, signed with HMAC or carrying a token.
- Your own code. The API starts an agent run or a workflow run, with an API key.
- Another agent. An agent can hand a task to your other agents, or to an outside agent over A2A.
How do you stay in control of what the agents do?
Every use case above runs on the same controls. Each agent gets only the actions you give it. Each action says whether it reads, writes or deletes. In the console, a new action that writes asks a person first until you decide otherwise, and anything that deletes always asks. The person is told in the console and by email, and the run waits for the answer.
Every tool call is recorded with its input and output, so you can read what an agent did and why. See human-in-the-loop approvals and AI agent security.
Which use cases does MyDeepAgents not cover yet?
Key points
- The most common first projects are support, inbox triage, CRM upkeep, reports, data questions and team chat.
- Pick work that repeats, follows written rules, starts by reading and is easy to check.
- Use a workflow when the steps never change, and an agent for the part that needs judgment.
- Agents start from a chat, a schedule, a webhook, the API or another agent.
- Approvals, narrow tool lists and a record of every call keep each use case under control.
Questions people also ask
What is the easiest AI agent to start with?
One that only reads and drafts: an inbox summary posted to Slack, or a ticket assistant that writes its suggested reply as an internal note. Nothing leaves without a person, and you learn how the agent behaves.
Can one agent do several of these jobs?
It can, but smaller agents are easier to test and safer to run. A common pattern is one agent per job, with a coordinating agent that hands tasks to the others.
Do I need to write code to set up these use cases?
No. Agents, workflows, connections and approvals are set up in the console. Code is only needed if you want to start runs from your own software through the API.
Which AI models can these agents use?
Models from the providers you connect with your own keys, such as Anthropic, OpenAI, Google and others. See how to choose a model for an AI agent.