AI Tasks let you run recurring work that requires judgment. You describe what you want the AI to do, set when it runs, and it works unattended: reading files, validating content, generating reports, screening uploads, auditing partner configurations, flagging anomalies, or verifying that offboarding was done correctly. A saved prompt supplies the instructions for each run.

The key difference from Automations is judgment. Automations follow rules. They move, copy, rename, and delete files the same way every time. AI Tasks read, assess, and act based on what they actually find. That makes them useful for work that depends on understanding what something says or means.

Each run creates a chat session you can review in Chat Logs. Site Administrators and Workspace Administrators can create and manage AI Tasks. Workspace Administrators can only manage tasks within their workspace. File contents and tool results used in a task can be processed by Files.com and its AI providers, AWS, Anthropic, and OpenAI, as part of carrying out the saved prompt. The same data-handling model applies to the interactive AI Assistant and unattended AI Tasks.

How AI Tasks Work

AI Tasks are agentic. When a task runs, the AI does not just generate a response. It reasons through a sequence of steps, takes actions, evaluates what it found, and decides what to do next. It can call multiple tools in a single run and chain those actions together based on what it actually finds.

Tasks run automatically on a schedule, when a file event occurs, or when an administrator triggers one manually. The saved prompt describes what to examine, the criteria to apply, and the report or changes to produce.

The AI figures out which tool to call at which step based on your prompt. You do not script the steps or configure each action individually. An Automation follows a fixed path. An AI Task follows a path that depends on what the AI finds.

Each step is recorded in a chat session you can review in Chat Logs, so you can see exactly what the AI did, which files it accessed, what decisions it made, and what it produced.

Permissions and Scope

An AI Task runs with its configured permissions within its owning site or Workspace. Full access is the default and supports administrative work as well as file operations. Files Only access limits the task to file and folder operations. The task does not use its creator's personal folder permissions as its access boundary.

Set the task's permission scope for the work it needs to perform. Use a Workspace when the task belongs within that Workspace's resources. A trigger folder or a path written in the prompt provides context; it does not restrict the task's permissions to that folder. Instructions in a prompt do not replace access controls.

Tasks run without asking for approval before each change. Review their permissions, prompt, and trigger together before enabling them. AI explains the relationship between task authority, provider processing, and controls for ending access.

What AI Tasks Can Do

AI Tasks run with the same capabilities as the AI Assistant. The AI can read files, analyze logs, check permissions and configurations, query user and group data, generate files, and take actions on your site based on what it finds.

When AI Tasks read files, they work best with text-based formats: documents, spreadsheets, CSVs, XML, JSON, HTML, and similar. They cannot read the visual content of images, audio, or video.

Content Assessment and Classification

When an intake folder receives documents with inconsistent names, a task can read their contents, identify the document type, and write a classification or review report. For example, it can distinguish an invoice from a shipping manifest without relying on the filename. A focused prompt can also ask it to identify missing invoice fields or passages containing restricted information.

These are AI assessments: an unflagged file is not proof that every required field is valid or every restricted value was found. Use deterministic checks, such as Expectations content validation for structured data, when the same rule must be evaluated consistently. An Expectation records the validation result; your workflow determines whether to release the file. Keep human review in the handoff when a judgment determines whether a document is suitable for release.

Configuration and Access Reviews

A scheduled task can compare partner-folder settings against a checklist stored on your site and write a report of differences. The checklist supplies the expected permissions, encryption requirements, or notification settings; the AI does not establish those requirements for you. Administrators can review the report before changing access or enabling a partner's workflow.

The same approach supports a review after user offboarding: name the user and the access areas to examine, and request a report of findings and checks that could not be completed. Review the underlying account and permission records before treating the report as confirmation that access has ended.

Activity Analysis

A task can combine delivery records and file information into a narrative report, such as which partner deliveries arrived during a specified period and which need investigation. Use Expectations for the arrival deadline itself, and Scheduled Report Exports when a built-in CSV report supplies the answer. AI adds value when the reader needs interpretation across those records rather than another copy of them.

When Not to Use AI Tasks

Files.com has purpose-built features that handle many common workflow needs without AI. Before creating an AI Task, check whether one of these already solves the problem.

Need	Use Instead
Moving, copying, renaming, or deleting files based on rules	Use Automations. They execute immediately, support retries, and do not involve the AI at all.
Renaming or organizing uploaded files automatically	The Renaming Uploaded Files and Organize Files folder settings handle this at upload time with no automation or task required.
Monitoring that expected files arrived on time	Expectations define what a correct delivery looks like and surface missing or late files without any AI.
Alerting on file, folder, or system events	Notifications cover file activity. Event Channels cover system events like Sync outcomes, Automation results, SSO activity, and user lockouts.
Synchronizing files between two locations	Sync replicates files between Files.com, remote servers, and cloud storage on a schedule or automation trigger.
Deleting files automatically after a set period	File Expiration removes files based on age with no AI or manual review.
Managing the lifecycle of user accounts	User Lifecycle Rules disable or delete accounts automatically based on inactivity. Use an AI Task only if you need a verification report after offboarding completes.
Encrypting or decrypting files automatically	GPG folder settings encrypt or decrypt files at upload with no prompt or task needed.
Generating scheduled reports on site activity	Scheduled Report Exports deliver built-in reports (Share Links, Permissions, Group Membership, Folder Size) as CSVs on a schedule. Use an AI Task only when you need narrative analysis or conclusions drawn from multiple sources.
Restricting what files can be uploaded	Data Governance restrictions block uploads that do not match your rules before files enter your environment. Use an AI Task only when you need to screen what is inside the file, not just its name or type.

Also avoid AI Tasks when:

The output must be identical every time. AI output can vary between runs. Use Automations when the result must be deterministic.
The files are images, audio, or video. AI Tasks cannot read visual or audio content.
A File Action trigger is on a high-volume folder with no source pattern set. Without a filename filter, the task runs on every file, including ones it cannot process, and can consume a lot of usage quickly.
The workflow requires a human to approve something mid-run. AI Tasks run unattended from start to finish.
Speed is critical. Automations execute in milliseconds. AI Tasks reason through each step before acting.
The workflow cannot tolerate a missed run. AI Tasks do not retry automatically. Automations have built-in retry logic for cases where reliability is required.

For a one-off task or a back-and-forth conversation, use the AI Assistant instead.

AI Tasks and Automations Together

Automations and AI Tasks work well together. Automations handle the reliable, rule-based parts of a workflow. AI Tasks handle the parts that require judgment.

For example, an Automation can move an incoming document to a staging folder, where an AI Task writes an assessment for review. A later Automation can route approved files using a fixed rule. The handoff must distinguish an AI assessment from the acceptance decision that releases a file.

Another: a scheduled AI Task reviews permission assignments across sensitive folders, identifies anything that looks unusual, and writes a report to an admin folder. A Notification then alerts the compliance team that the report is ready to review.

You can combine AI Tasks with Event Channels and Notifications to close the loop between unattended AI work and the people who need to act on it.

For details on editing prompts and triggers, viewing run logs, and troubleshooting failed runs, see Managing Tasks.