Understanding AI Agents

The AI agent in Agency Hero is your always-available assistant — one that knows your clients, your meetings, your open tasks, and the decisions your team has made. It isn’t a generic chatbot you have to explain everything to every time. It’s a context-aware workspace companion that draws on everything you’ve built up in Agency Hero to give you answers and take action on your behalf.

This article explains what agents are, how they work, and what you can ask them to do.

What Is an AI Agent?#Copy link

At its core, an agent is a coordinator — it listens to what you ask, figures out the best way to help, gathers the relevant information, and either answers your question or takes the action you need.

The agent is not a single fixed chatbot. It’s a runtime that dynamically loads:

• The right skill for your task (more on skills below)
• Your workspace context — client details, meeting history, decisions, open tasks, risks
• The tools it needs — search, task creation, CRM lookup, and more
• Your permissions — so it only does what it’s allowed to do

This means the same agent can help you prep for a client call, pull a list of open risks, write a memo, or create a follow-up task — adapting to what you need in that moment.

How the Agent Understands Your Workspace#Copy link

Every workspace in Agency Hero accumulates context over time: meeting summaries, extracted decisions, open questions, tracked topics, tasks, documents, and intelligence items. The agent taps directly into all of this when you ask a question.

Four Knowledge Sources#Copy link

The agent searches across four separate knowledge backends — and it knows which one to use for each type of question:

For broad questions like “What do we know about DevOps?” the agent searches all four sources in parallel and synthesizes the results. For targeted questions like “What decisions have we made about pricing?” it goes straight to the intelligence registry.

Context Pack#Copy link

Before answering, the agent loads a context pack — a pre-assembled snapshot of the workspace’s current state. This includes things like open tasks, tracked topics, recent activity, and key stakeholders. Think of it as the agent’s briefing document for your workspace. For live questions about current state (“what are our open risks?”), the context pack gives instant answers without a full search.

Workspace Boundaries#Copy link

The agent always respects workspace boundaries. If you’re in the Acme Corp workspace, the agent queries Acme Corp data by default. It won’t mix in data from another client unless you explicitly ask for cross-workspace results. When you ask “show me tasks across all my workspaces”, the agent fans out to every workspace you have access to — but each search is scoped and results are clearly attributed by workspace.

What You Can Ask the Agent to Do#Copy link

Here’s a practical sense of what the agent handles well:

Search and Look Things Up#Copy link

• “What were the key decisions from last week’s Acme meetings?”
• “Find all open risks in this workspace”
• “Search for anything we discussed about the API migration”
• “What topics are we tracking in this workspace?”
• “Show me the latest meeting summary”
• “What do we know about the Q3 budget?”

Task Management#Copy link

• “Create a task to follow up with Sarah about the contract by Friday”
• “Show me all open tasks in Acme Corp”
• “What’s on my plate this week across all my workspaces?”
• “Mark the onboarding task as complete”

Meeting Preparation and Debrief#Copy link

• “Prep me for tomorrow’s Acme quarterly review”
• “Quick context for my call starting now”
• “What were the action items from Monday’s planning meeting?”

Generate Content and Memos#Copy link

• “Draft a memo summarizing our decisions about the redesign project”
• “Write a project status update based on this week’s meetings”
• “Help me write a client email about the Q4 timeline”

Workspace and Portfolio Views#Copy link

• “Give me a status update across all my client workspaces”
• “What’s my week look like?”
• “Which clients have had no meetings in the last two weeks?”

Research and Deep Dives#Copy link

• “Research everything we know about our AI strategy — I’ll check back after lunch”
• “Summarize all discussions about pricing across our client workspaces”
• “What are the recurring risks showing up across our accounts?”

Skills: How the Agent Adapts to Different Tasks#Copy link

Not every task requires the same approach. A meeting prep request needs different tools, context, and behavior than a deep research task. Agency Hero handles this through a skill-based architecture.

What Is a Skill?#Copy link

A skill is a focused set of instructions and permissions that tells the agent how to behave for a specific type of task. Each skill defines:

• The role the agent takes on (e.g., “strategic account analyst” for QBR preparation)
• Which tools it’s allowed to use
• What guardrails apply (e.g., no external writes without approval)
• What context it needs to do its job

When you start a conversation, the agent automatically identifies which skill fits your request and loads it. If you ask about meeting prep, it loads the meeting-prep skill. If you ask a general question, it uses the general assistant skill as the default.

System Skills#Copy link

Agency Hero ships with built-in system skills:

• General Assistant — the default; handles everything from search to task creation
• Meeting Prep — deep preparation for upcoming meetings, including context, open items, and talking points
• Task Manager — creates, updates, and organizes tasks
• Research Assistant — in-depth research requiring multiple search iterations or web lookups
• Doc Author — writes and improves documentation articles
• And more, with additional skills being added regularly

You can also select a skill manually by typing @ followed by the skill’s slug in the chat input. For example:

• @research-assistant — directs the agent to use the Research Assistant skill
• @doc-author — switches to the Doc Author skill for writing and editing documentation
• @meeting-prep — loads the Meeting Prep skill for call preparation

This is useful when you want to direct the agent to a specific mode without relying on it to infer the right skill from your message.

Specialized Subagents#Copy link

For complex or time-consuming tasks, the agent can spawn a subagent — a separate, focused assistant that works on a specific problem in parallel. For example, when you ask for a status update across five client workspaces, the agent can run five simultaneous workspace lookups and synthesize the results, rather than doing them one at a time.

Each subagent:

• Has its own isolated context (so complex tasks don’t slow down your main conversation)
• Inherits the same permissions as the parent agent — it can never do more than you’re allowed
• Returns its results to the main agent, which synthesizes and presents them to you

You generally won’t notice subagents working in the background — the experience is just a faster, more thorough answer.

Operating Across Workspaces#Copy link

Agency Hero is designed for agencies managing multiple clients at once. The agent reflects this with first-class support for cross-workspace work.

By default, the agent operates in your current workspace. If you’re in Acme Corp, it searches Acme Corp data, creates tasks in Acme Corp, and responds in the context of that client relationship.

For cross-workspace queries, just say so explicitly:

• “Show me tasks across all my workspaces”
• “What’s happening across all my clients this week?”
• “Which workspace has the most overdue tasks?”

When you refer to a specific workspace by name — “show me meetings in the Vertice Labs workspace” — the agent targets that workspace directly, even if it’s not the one you’re currently viewing.

When results span multiple workspaces, the agent always groups them by workspace name so you know exactly where each piece of information comes from. No data from one client ever leaks into another.

How the Agent Keeps You in Control#Copy link

Read First, Write Second#Copy link

The agent defaults to reading and searching. When it needs to take an action that changes data — creating a task, saving a memo, updating a record — it tells you what it’s about to do and confirms before proceeding.

For actions that could affect external parties (sending an email, posting to a client Slack channel), approval is always required. You’ll see a preview of what will be sent before anything goes out.

Tool Approvals#Copy link

Some agent actions require your explicit approval before they execute. This is determined by a risk tier system:

• Read-only actions (searching, listing, looking things up) — happen immediately, no approval needed
• Internal writes (creating a task, saving a memo) — generally execute automatically
• External or higher-impact writes (sending an email, updating CRM data) — pause and ask for your approval first

You’ll see an approval card with a summary of what the agent wants to do, and you can approve, edit, or cancel.

Workspace Permissions#Copy link

Agency Hero admins can configure what the agent is allowed to do within a workspace — for example, restricting certain tools or requiring approval for specific actions. These settings are enforced at every step, so the agent can never bypass the rules set for your organization.

Practical Tips#Copy link

Be specific about scope. If you want results from a specific client, name them: “show me decisions from Acme Corp”. If you want everything, say “across all my workspaces”.

Ask follow-up questions. The agent remembers the conversation context. If it surfaces five decisions, you can say “expand on the third one” without repeating yourself.

Use it for recurring briefings. Asking “what’s my week look like?” every Monday morning is a great habit — the agent pulls together your meetings, tasks, and open items across all your clients into a single summary.

Let it capture intelligence. When a decision, risk, or open question comes up in conversation, you can tell the agent to record it: “capture that as a decision” or “log this as a risk”. It will add a structured item to the workspace for future reference.

For long research tasks, come back later. If you ask for a deep research task — like summarizing everything across multiple clients on a topic — the agent can run it in the background. Say “have it ready when I’m back” and the results will be waiting for you.

What the Agent Won’t Do#Copy link

The agent is designed to be helpful and safe, not to take unilateral action. A few boundaries to know:

• It won’t send anything to a client without your approval.
• It won’t invent facts — if it’s not sure, it says so and offers to search.
• It won’t access data you don’t have permission to see.
• It won’t cross workspace boundaries unless you ask.

These aren’t limitations — they’re features. When you’re managing multiple client relationships, you want an assistant that respects those boundaries by default.