download dots

🎛️ Agent Answer With a Model Escalation with AI

A comment is answered by your named agent on its default model, and only the questions it marks as hard are re-run on a model you picked yourself.

  • Event-triggered
  • 100+ integrations
  • Branching logic
  • Schedule + monitor
UNIT · TASKADE.FLOW3 STEPS

How this flow runs

When

When someone comments

Taskade

Then 1

Agent

Project Assistant

Ask an agent · Taskade AI

Knows
every project, document and thread in the workspace
Remembers
the questions it has already answered here

Then 2

Branch

Branch

Take the first path whose condition is true, each with its own steps. Paths are evaluated in order.

  1. 01Answer is ready

    Add a comment

    Taskade

  2. 02Needs a specialist model

    Ask Claude

    Anthropic Claude

    Add a comment

    Taskade

This agent carries its knowledge and memory between runs, and can call your other automations as its own tools. How custom agents work

Branch is a step you add in the builder, in the same list as an app step — there is no separate scripting surface. All automation actions

Build this flow with AI

Opens the builder with these steps already described. Free to start, no card.

What does this automation do?

A comment is answered by your named agent on its default model, and only the questions it marks as hard are re-run on a model you picked yourself.

What can you do with it?

  • Keep the default model for the answers it already handles well
  • Escalate the hard question rather than the whole workload
  • Post both kinds of answer in the same place, on the task

The Steps

This automation runs in 3 steps:

  1. When Taskade — When someone comments
  2. Then Taskade AI — Ask an agent ("Project Assistant")
  3. Then Branch — Answer is ready: Add a comment · Needs a specialist model: Ask Claude, Add a comment

What makes this different from a Zapier-style flow?

The "Project Assistant" step is not a one-off model call. It is a Taskade AI agent that knows every project, document and thread in the workspace, remembers the questions it has already answered here, so every run builds on the last one instead of starting from an empty prompt.

A stateless "ask the model" step forgets everything the moment it returns. An agent keeps its knowledge and its memory between runs, carries its own tools, and can be handed work by other automations.

Who is this automation for?

Built for teams that want one agent but not one model for everything. A comment is answered by your named agent on its default model, and only the questions it marks as hard are re-run on a model you picked yourself.

How do you set it up?

  • Open the "Automations" tab in your workspace.
  • Click ➕ Add automation, then start from scratch.
  • Add the trigger "When someone comments" from Taskade.
  • Add the action "Ask an agent" from Taskade AI. Point it at an agent called "Project Assistant".
  • Add a Branch step: Branch — Answer is ready: Add a comment · Needs a specialist model: Ask Claude, Add a comment.
  • Enable the toggle in the top-right corner to activate the automation.

Prefer to describe it instead? Click Build this flow with AI above and Taskade wires the trigger, the steps, and the agent for you.