Automate with Agents and Workflows

Automate with Agents and Workflows

Outrun gives you three building blocks for automation. Everything you build is a combination of these primitives. Understanding when to use each one is the key to getting value quickly.

The Three Primitives

Agent

One task, one AI.

An Agent is a single AI specialist focused on one job. You give it a role, connect it to a knowledge base, and point it at a communication channel.

Use an Agent when:

• The task is self-contained and does not require coordination with other specialists
• You need a single point of contact for a specific function
• The job can be described in one sentence

Examples:

• "Answer support questions using our product docs"
• "Qualify inbound leads based on our ICP criteria"
• "Summarize new Confluence pages into Slack updates"

An Agent works on its own. It receives input, thinks, and responds. No handoffs, no routing, no multi-step logic.

Agent Group

Multiple specialists collaborating.

An Agent Group is a team of Agents that work together on a shared task. Each Agent in the group has a different specialty. Outrun routes incoming requests to the right specialist automatically, and Agents can hand off to each other when a conversation crosses domains.

Use an Agent Group when:

• The problem space is too broad for one Agent to handle well
• Different queries need different knowledge bases or expertise
• You want automatic routing without building manual rules

Examples:

• A pre-sales team: a qualifier Agent, a technical specialist, and a solutions architect
• A support team: a billing Agent, a product Agent, and an escalation Agent
• A research team: a market analyst, a competitor tracker, and a summarizer

Agent Groups handle the routing for you. You define the specialists and their expertise areas, and Outrun figures out who should handle each request.

Workflow

Multi-step process with logic.

A Workflow is a sequence of steps that runs when something happens. Steps can include AI processing, conditional logic, human review gates, API calls, and notifications. Workflows are event-driven: they start when a trigger fires (a webhook, a schedule, a data change).

Use a Workflow when:

• The process has multiple steps that must happen in order
• You need conditional logic ("if X, then do Y, otherwise do Z")
• You want a human-in-the-loop approval step before an action is taken
• The automation should run on a schedule or in response to an event

Examples:

• "When a deal goes stale for 7 days, AI drafts a follow-up email, sends it to the account owner for review, then sends it"
• "Every Monday, summarize last week's support tickets and post the summary to Slack"
• "When a new lead comes in, enrich from our CRM, score it, and route high-value leads to the sales team"

Workflows give you control over the order of operations, branching, and human oversight.

Choosing the Right Primitive

Combining Primitives

The real power comes from mixing these together. Here are some patterns.

Pattern: Agent Group + Workflow Follow-Up

A pre-sales Agent Group handles live chat on your website. When the conversation ends, a Workflow picks up: it logs the interaction to your CRM, scores the lead, and notifies the assigned sales rep if the lead is high-value.

• Agent Group handles the real-time conversation (qualification, technical questions, pricing)
• Workflow handles the post-conversation automation (CRM update, scoring, notification)

Pattern: Workflow with AI Steps

A Workflow triggers every morning on a schedule. It pulls yesterday's closed-lost deals from your CRM, passes each one to an AI step that drafts a "what went wrong" summary, then posts the batch to a Slack channel for the sales manager.

• Workflow orchestrates the schedule, data pull, and delivery
• AI nodes within the Workflow do the summarization

Pattern: Agent for Triage, Workflow for Action

A single Agent monitors your support inbox and categorizes incoming tickets by urgency. High-urgency tickets trigger a Workflow that pages the on-call engineer, creates a Jira ticket, and sends an acknowledgment to the customer.

• Agent handles the classification (one task, one AI)
• Workflow handles the multi-step response (page, create ticket, send email)

Pattern: Data Change Triggers Agent Group

A Workflow listens for new contacts added to your CRM. When a new contact arrives, the Workflow checks if they match your ICP criteria. If they do, it hands the contact off to an Agent Group that runs a personalized outreach sequence across email and LinkedIn.

• Workflow handles the trigger and filtering logic
• Agent Group handles the multi-channel outreach (different specialists for different channels)

Getting Started

1. Start with One Agent

Pick your most common repetitive task and build a single Agent for it. Connect it to the relevant knowledge base and let it run for a few days. This teaches you how Agents think and where they need guidance.

2. Add a Workflow for Oversight

Once your Agent is handling tasks reliably, build a Workflow around it. Add a human-in-the-loop step so you can review the Agent's work before it takes action. This is especially important for anything customer-facing.

3. Expand to Agent Groups

When you notice that one Agent is being asked questions outside its specialty, create an Agent Group. Split the responsibilities across multiple specialists and let Outrun handle the routing.

4. Connect Your Triggers

Set up data change triggers and schedules to make your automation proactive. Instead of waiting for someone to ask, let Outrun watch for events and act on them automatically.

Key Concepts

Knowledge Bases

Agents and AI Workflow nodes pull context from Knowledge Bases. A Knowledge Base is a collection of documents (synced from Confluence, uploaded manually, or pulled from your sources) that gets chunked, embedded, and made searchable. See Connect Your Sources for how to get data into Outrun.

Human-in-the-Loop (HITL)

Any Workflow can include a HITL node that pauses execution and waits for a human to approve, edit, or reject the AI's output before the next step runs. Use HITL for:

• Customer-facing communications (emails, chat responses)
• Financial or legal actions
• Anything where a mistake would be costly

Triggers

Workflows start with a trigger:

• Webhook -- An external system sends data to Outrun's webhook URL
• Schedule -- A cron expression runs the Workflow on a recurring basis
• Data change -- A record is created or updated in your connected sources

Style Guides

You can attach a Style Guide to any Knowledge Base. This teaches your Agents to match your brand voice, terminology, and communication standards.

Next Steps

Need help? Contact us at [email protected] or join our Discord.