AI Agent can do more than summarize your data. It can also propose actions, such as updating a CRM field, creating a task, drafting an email, or using a connected MCP tool.
Pod keeps you in control of those changes. When AI Agent wants to write, send, create, or update something, Pod shows you an approval request first. Nothing is completed in the connected tool until the action is allowed.
This is where agents stop being "another place to ask questions" and start being a real force multiplier. Insights and answers are useful, but the time-consuming part of a seller's day is the work that comes after the insight — the CRM update, the follow-up email, the prep deck, the task list. When your agents do that work in the background, you get leverage: more pipeline moved per hour, fewer things falling through the cracks, and more time for the parts of the job that actually need you in the room.
This article covers what AI Agent actions are, where approval requests appear, how to allow or reject them, and what to do when an action fails.
What is an AI Agent action?
An AI Agent action is a step AI Agent wants to take in one of your connected tools.
Examples include:
Drafting or sending an email.
Updating a CRM field, such as close date, stage, or next step.
Creating a follow-up task.
Logging a call or adding a note to a record.
Using a connected MCP server to work in another system.
Actions can come from a manual AI Agent run or from a Skill that runs automatically through an Automation. In both cases, the action belongs to the AI Agent runtime. The Skill or prompt tells AI Agent what to do, and AI Agent proposes the action for review.
Read actions vs. write actions
Not every tool call needs approval.
AI Agent can use two broad types of tools:
Read tools fetch information, such as CRM details, meeting context, or connected system data. These usually run automatically.
Write tools change something, such as creating, updating, sending, or deleting data. These require review unless your workspace has an approved policy that allows that specific write automatically.
When a write action needs review, Pod shows an action card with the details AI Agent plans to use.
Where approval requests appear
Approval requests appear where you already work with AI Agent.
You may see them:
In an AI Agent conversation in the Pod web app.
In an AI Agent conversation in the Chrome Extension.
On Agent Feed cards for automated runs.
If an automated run is also sent to Slack, the Slack message links back to Pod. Approval happens in Pod, not inside Slack.
Manage an action in an AI Agent conversation
When AI Agent proposes an action during a conversation, Pod adds an action card to the thread.
To review it:
Open the action card.
Review the destination, tool, target, parameters, policy, and status details.
Choose Allow once, Always allow, or Reject.
Watch the status update as Pod runs or resolves the action.
Continue the conversation once you are done.
You can review multiple action cards in any order. You do not need to resolve them in the exact order AI Agent proposed them.
Allow once
Use Allow once when you want Pod to run this specific action one time.
This is the safest default for actions you want to inspect individually, especially CRM updates, emails, and writes through MCP tools.
Always allow
Some actions may show Always allow.
Use it only when you are comfortable letting Pod run matching future writes without asking each time. This is useful for low-risk, repeatable actions, but it should be used carefully for anything customer-facing or data-sensitive.
If Always allow is unavailable, Pod explains why. You can still use Allow once for the current action.
Reject
Use Reject when you do not want the action to run.
AI Agent can continue the conversation after a rejection. If you want a different version of the action, tell AI Agent what to change in your next prompt.
Manage actions from Agent Feed
For automated runs, action approvals can appear directly on Agent Feed cards.
From Agent Feed:
If a card has one or two pending actions, each action appears directly on the card.
Feed action buttons use hold-to-confirm, so you need to press and hold before the action is allowed or rejected.
If a card has more than two pending actions, click Review actions to open the full AI Agent run.
When every action on a card is resolved, the card shows the resolved action state.
Agent Feed action controls are available in both the Pod web app and the Chrome Extension.
For more on reviewing automated results, see Use the Agent Feed.
Action statuses
An action moves through a short set of statuses.
You may see:
Pending — AI Agent proposed the action and is waiting for review.
Executing — you allowed the action and Pod is running it.
Executed — the action ran successfully.
Failed — Pod tried to run the action, but the connected tool rejected it or returned an error.
Rejected — the action was rejected.
Auto-rejected — Pod closed the pending action automatically before continuing.
Stale — the action has been pending long enough that approval-time revalidation may fail.
Allowed by policy — the action ran automatically because a saved policy allowed it.
What Stale means
Pending actions can become stale when they have been waiting for a while.
When an action is stale, the data AI Agent planned to use may no longer match the latest state of the record or connected system. You can still allow the action, but Pod may need to revalidate it before it runs.
If revalidation fails, the action moves to Failed with a reason.
What happens if you continue before resolving an action
If you continue the conversation while an earlier action is still pending, Pod may close the unresolved action before the next run continues.
This prevents old approval requests from blocking the conversation. AI Agent can still propose a new or corrected action later.
When an action fails
If an action fails, the card shows Failed with a short reason.
Common failure reasons include:
Unauthorized — Pod does not have permission to run this action. Check the connected integration or ask a workspace admin.
Tool disabled — an admin disabled the tool. Ask an admin to review the connected server or tool settings.
Server disconnected — the connected server is offline or was removed. Reconnect it before trying again.
Auth expired — the authorization for the connected tool expired. Reconnect the tool or MCP server, then try again.
Invalid payload — the action parameters did not match what the tool expected. Ask AI Agent to try again with different details.
Timeout — the connected tool did not respond in time. Try again, or check the connected service if the issue continues.
Execution error — the connected tool returned a general error. Try again or contact support if it keeps happening.
Policy denied — a saved policy blocked the action.
Policy unavailable — Pod could not apply the saved policy for this action.
For MCP-related failures, admins can review connected servers in MCP settings. For setup help, see Connect MCP servers.
Who can manage an action?
Action controls are shown to the user who owns the run.
In general:
If you start an AI Agent run manually, you can allow or reject actions from that run.
If an Automation runs a Skill on your behalf, the actions appear in your Agent Feed.
If you manage another user's feed, you may be able to review their automated results depending on your workspace access.
If you open a run where you do not have approval access, you may still see action details and statuses, but the allow/reject controls are hidden.
Related articles
Use Launch and use AI agents to learn how to start a run, add Skills, and write prompts.
Use How AI Agents think and work to understand reasoning, tool calls, and the Thoughts panel.
Use Use the Agent Feed to review automated AI Agent results and pending actions.
Use Connect MCP servers to give AI Agent access to more external tools.
Use Set up Skill automations to run Skills automatically.
Need help? Send us a message via the in-app chat or email us at [email protected].
Want to talk to someone? Book a session with one of our specialists.