AI agents in workflows

Most workflow steps do one fixed thing: update a field, create a task. The two AI steps are different: you give them an instruction, and they reason over the data flowing through the workflow. This is how you automate judgment, not just plumbing: classifying leads, summarising context, deciding what a record needs next.

Both live in the AI Agent group of the action picker.

AI Agent

The AI Agent step takes a prompt and returns output for later steps.

Configuring it:

1. Prompt: describe what the AI should do, using variables from the trigger and earlier steps (e.g. {{trigger.record.name}})
2. Business context: attach business context files so the agent knows your product, positioning, and rules
3. AI Response Schema: define output fields (name + a brief explanation of each) that force the AI to answer in a fixed structure instead of free text

The schema is the key feature: each output field becomes a variable that later steps can reference. An agent that returns { priority, reason, suggestedAction } can feed an If/Else on priority, write reason into a CRM field, and put suggestedAction into a task body.

Example: inbound lead triage:

• Trigger: Record is created (Person)
• AI Agent: "Classify this lead against our ICP" with outputs fit (high/medium/low) and why
• Switch on fit → high: add to priority sequence + task for the owner; medium: nurture sequence; low: update a status field

AI CRM Agent

The AI CRM Agent goes further: it can read and act on your CRM directly. Instead of only returning text, it works against a record you point it at.

Configuring it:

1. Prompt: what the agent should do
2. Entity / Pipeline: which object the agent operates on
3. Record: the specific record (usually a variable from the trigger)
4. Options:
   • Enable thinking: deeper reasoning for harder tasks, with a configurable thinking effort
   • Enable web search: let the agent research on the open web while working

The step's run detail shows the input, the output, and the agent's reason, so you can audit what it did and why, visible in the workflow run logs.

Example: account research on deal creation:

• Trigger: Pipeline record is created (Opportunity)
• AI CRM Agent on the linked Company: "Research this company's recent news and update the Research Notes field; enable web search"
• Create Task for the owner: "Review AI research before first call"

When to use which

	AI Agent	AI CRM Agent
Returns	Structured output (your schema)	Actions on CRM data + output
Touches records	No (later steps do that)	Yes, directly
Best for	Classification, drafting, extraction feeding other steps	Research-and-update, record-level judgment calls
Extras	Business context files, response schema	Thinking mode, web search

Rule of thumb: if you want to keep deterministic control of what changes in your CRM, use AI Agent and let explicit Update Record steps apply its output. If you want the agent to handle the whole job on a record, use AI CRM Agent.

Cost awareness

AI steps consume credits per execution. The Workflows Overview has a Token consumers panel showing which workflows use the most; check it before scaling an AI-heavy workflow to high-volume triggers, and put Filter steps before AI steps so you only pay for records that matter.

Key outcome

AI agent steps let workflows make the calls that used to need a human (qualify, research, summarise, decide) while the rest of the workflow applies the results with normal deterministic actions. That combination is what turns automation from data plumbing into an actual working teammate.