AI and agents
Esca Automations uses AI in two ways: to help you build automations, and as steps inside an automation that hand work to an AI agent. Agent steps can also pause for a person to review or approve what the agent proposes.
For field-level detail on the AI actions, see the Reference.
Building with AI
The Build page turns a plain-language description into a flow. Describe what you want—for example, "When an article is published, summarize it and post the summary to Slack"—and Esca drafts an automation with the triggers, conditions, and actions to match. Review and adjust the result in the editor, then save it like any other automation.
The AI Agent action
The AI Agent action runs an AI agent as a step in the flow. You give it a prompt describing the task, written with variables so it can work on the trigger's data and earlier results, and one or more MCP servers whose tools the agent may use. The agent runs on the automation server itself, so it can combine tools from several environments in a single run—reading from one Brightspot instance and writing to another, for example. An optional template—a reusable agent from the Agents catalog—steers how the agent approaches the task, and an optional output shape makes it return structured data that later steps can read, alongside its message.
The Remote AI Agent action
The Remote AI Agent action hands the task to a connected Brightspot CMS, whose own AI agent does the work using that CMS's tools. Choose the connection, write the prompt, and optionally pick an author persona from the connected CMS whose writing style the agent adopts. Use it when the task belongs entirely to one CMS; use the AI Agent action when the agent needs tools from more than one system.
The AWS AgentCore action
The AWS AgentCore action invokes a hosted AWS Bedrock AgentCore agent. Pick a hosted agent from your AWS environment's catalog—separate from the Agents catalog below—supply its input as a prompt or as JSON, and choose how much it should remember between runs—from this run only, up to everything for the site. As the agent works, its progress streams into the run.
The AI Judgment condition
AI can route a flow as well as act in it. The AI Judgment condition asks an agent a yes/no question about the data passing through—"Is this on-brand?"—and routes to true on yes and false on no. Like the AI Agent action, it can draw tools from MCP servers and be steered by an agent from the catalog, but it is only ever offered read-only tools: deciding a route never changes anything. The agent's reasoning is written to the step's log, so you can see why a run went the way it did. See Conditions and branching and the Reference.
Human-in-the-loop
An agent can pause mid-run to ask a person a question—to confirm a decision, choose among options, or supply missing detail. When it does, the flow waits on that step and the request is presented for someone to answer, for example in Slack. Once the person responds, the agent resumes with their answer and continues. This lets you keep a human in control of consequential steps while the agent does the rest.
Pausing is opt-in per step. Enable it with the agent step's Human in the Loop setting and choose the providers the question is pushed to—for example, Slack; every request is also answerable from the AI Approvals inbox. A step without the setting never pauses: the agent must finish the task on its own.
The AI Approvals inbox
The AI Approvals page lists every step across every automation that is currently waiting on a person, in one place—so you don't have to know which automation paused to find and answer it. Each row shows the automation, the run, the step, and how long it's been waiting; Respond opens the same answer form the request's notification (for example, in Slack) links to.
You can also subscribe to be notified in the CMS whenever a run you're watching pauses for input, from your profile's notification settings, scoped to specific automations if you like.
The Agents catalog
The Agents page is the catalog of reusable agents that steer AI steps: pick one in the Template field of the AI Agent action or the AI Judgment condition, and its guidelines shape how the agent approaches the task. The catalog merges the built-in agents, which are read-only, with custom agents you create on the page itself. Browse it to see what each agent does before wiring one into a flow.