# What Agents Can Do > Source: https://docs.trailspark.ai/docs/what-agents-can-do ## Overview An agent connected through [Agent Access](/docs/agent-access-overview) can do much of what you do in the app — as long as you've granted the matching permission. This page walks through the most common use cases and shows which of the three permissions — **Read**, **Configure**, **Data corrections** — each one needs. Everything below is scoped to the single workspace the agent was granted access to, and account re-evaluation stays off unless you've explicitly allowed it (and set a daily limit). ## Review accounts, leads, and evaluations > Needs **Read** An agent can find and read your scored accounts and the people in them: - Search leads and open a single lead's full profile - Browse target accounts and open a single account - Read evaluation results and history — the score, and how it's changed over time This is the foundation for reporting, triage, and "tell me what's happening with account X" questions. See [Browsing Accounts](/docs/browsing-accounts) for the same information in the app. ## Review cleaned signal activity > Needs **Read** An agent can list the **cleaned, customer-facing signal activity** for a specific lead or a specific account, newest first. These are the same mapped signals you see in the app — page views, form submissions, product usage, and so on — not raw webhook payloads. Activity is always anchored to one lead or one account at a time, so an agent reviews a specific entity's timeline rather than pulling a workspace-wide firehose. A typical flow is: search for a lead or account, then read that entity's recent signal activity to explain *why* it's scoring the way it is. ## Build and refine ICPs > Needs **Configure** An agent can do the same ICP work you'd do in the ICP builder: - Create a new ICP or update an existing one - Adjust **eligibility** rules (which accounts match the ICP) and **scoring filters** (what's worth scoring) - Manage buying-group **roles** and their point weightings - Preview the impact of a change before committing, and — if you've allowed re-evaluation — re-score affected accounts See [ICP Creation](/docs/icp-overview) for how these pieces fit together. > [!NOTE] > Re-scoring accounts is billable, so it only happens when you've turned on re-evaluation for that key or connected agent, and it stays within the daily limit you set. ## Connect a CRM or other integration > Needs **Configure** An agent can check the status of your integrations and **start** a CRM connection — but a person still finishes it. Rather than connecting silently, the agent hands you off to the provider's own sign-in and approval screen (for example, HubSpot or Salesforce), so the connection is always completed by someone who can authorize it. See [CRM Integration](/docs/crm-integration-overview) for the connections TrailSpark supports. ## Map incoming signals > Needs **Configure** An agent can manage how raw incoming events become the cleaned signals you score on: - List and read your signal-mapping rules - Create or update a rule, and activate or deactivate it - Preview what a rule would match, and check the impact before turning it on This is the same signal-mapping work described in [Signals Overview](/docs/signals-overview) — useful for keeping mappings tidy as your sources change. ## Correct data and give feedback > Needs **Data corrections** When an agent spots something wrong, it can help fix it: - Correct a person's buying-group **role** on an account - Submit feedback on an evaluation to help improve future scoring These corrections feed the same learning loop as corrections you make by hand in the app. ## Permissions at a glance | Use case | Permission | |---|---| | Review accounts, leads, evaluations | Read | | Review cleaned signal activity | Read | | Inspect ICPs, mappings, integration status | Read | | Build and refine ICPs | Configure | | Connect a CRM (start the flow) | Configure | | Map incoming signals | Configure | | Manage destinations | Configure | | Correct roles and give evaluation feedback | Data corrections | Grant only the permissions an agent actually needs — see [Agent Access Overview](/docs/agent-access-overview#permissions) for the full permission model, and [Managing Agent Keys](/docs/agent-keys) or [Connecting an AI Agent](/docs/connect-an-agent) to set one up.