What Is an AI Agent in Decision Terrain?

An agent is a persistent AI employee that can take on a role, keep context, use tools, work with files, communicate through channels, and ask for approval when it needs a decision or sensitive access.

Think of an agent as a named teammate with a job description, not as a one-off chat. You can create one for research, operations, sales prep, content workflows, customer support, monitoring, internal reporting, or another recurring responsibility.

The main pieces

Charter

The charter is the agent's standing instruction. It describes the role, goals, boundaries, tone, cadence, and any systems or contacts the agent should use.

A good charter says what outcome matters and what the agent should avoid. For example:

Prepare concise weekly competitor updates for the product team. Use files in the workspace as background context. Do not send external messages unless I approve them.

Timeline

The timeline is the working record for an agent. It can include messages, plan updates, deliverables, tool activity, files, pending requests, and results.

Use the timeline to understand what happened, what the agent is doing now, and where it needs input.

Plan and deliverables

For larger requests, the platform may create or update a plan. The plan turns a broad goal into visible steps and deliverables so you can steer the work before too much effort is spent.

See Planning And Deliverables.

Filespace

Each agent can have a workspace for reusable files. Files can come from chat uploads, email attachments, generated outputs, or direct file management.

See Files And Workspaces.

Channels and contacts

agents can communicate through web chat, email, approved contacts, app channels, and inbound webhooks depending on how they are configured.

See Channels And Contacts.

Tools and apps

Tools give an agent capabilities beyond normal chat. Examples include file operations, browser work, HTTP requests, connected apps, MCP servers, built-in communication tools, and configured system skills.

See Tools And Apps.

Approvals

Approvals keep humans in control. An agent can ask for input, request credentials, ask to contact a person, or propose creating or changing another agent.

See Approvals And Requests.

the platform, agent, task, and template

Use these terms consistently:

Term	Use it for
Agent	The end-user AI employee you create, chat with, configure, and manage.
Agent	The developer/API term for the same underlying persistent resource.
Task	A smaller unit of work. End users usually do not need to manage tasks directly.
Template	A reusable starting point for creating an agent with a known purpose.

Legacy API and implementation areas may still use terms such as browser-use task or browser automation. Those terms describe a technical surface, not the whole product.

What an agent is good at

agents are best when the work has a clear role, inputs, boundaries, and a useful output. Strong examples include:

• Monitoring a market, vendor, inbox, queue, or dashboard.
• Preparing research briefs or recurring summaries.
• Turning files and notes into organized deliverables.
• Coordinating work across tools or connected apps.
• Handling repetitive operations with human approval checkpoints.
• Keeping a team informed through chat or email.

What to define up front

Before creating an agent, decide:

• What job should this agent own?
• What information can it use?
• Which tools or apps should it access?
• Which people or channels can it contact?
• What should require approval?
• How will you review inference usage and decide when to pause work?
• What should count as a finished deliverable?

You can refine all of this later. The first version only needs enough context for the platform to begin safely.