Platform · Sub-Agents

One agent, with a team behind it

When a task needs depth, an Idam AI agent spins up focused sub-agents that work in parallel, each handling one angle, then merges what they find into a single clear answer. You get the thoroughness of a team without running the meeting.

In parallel
depth without the wait
Every angle
fewer blind spots
Traceable
see which angle flagged what
One agent, four specialists, in parallelworking
PRD ReviewerLegalTechnicalSalesDesignOne merged review
The Problem

A single pass has blind spots

Hard problems need more than one perspective. A lone agent doing everything at once tends to go wide and shallow. The depth you want takes a team, and assembling that team by hand is its own chore.

One pass, one angle

A single model making one pass tends to see what it looks at first. The legal risk, the edge case, the thing engineering will ask about: easy to miss when nobody is assigned to look for it.

Depth costs you prompts

Want a second perspective? You ask again. A third? Ask again. Getting thorough output means running the same task five times with different instructions and stitching the answers yourself.

Thorough means slow

Reviewing a draft from four angles one after another is four times the wait. So you settle for one angle and hope the rest holds up.

What Sub-Agents Do

Depth on tap, no extra prompts

You ask once. Behind the scenes, the agent fans the work out to specialists and pulls their answers back together. This is depth inside one agent. For coordination across several agents, see multi-agent orchestration.

PRD Reviewer4 sub-agents, in parallel
Legal
Passed
Technical
2 flags
Sales
Passed
Design
1 flag
Merged into one review · 3 flags to address

A team forms around the task, then dissolves

Sub-agents exist for the length of the task and nothing more. They spin up when the work calls for it and wind down once they have reported back, so you get the depth without managing a roster.

  • The agent decides when a task needs more than one pass
  • It spawns focused sub-agents, each with a single job
  • They run at the same time, not one after another
  • Their findings merge into one answer you can act on
The Specialists

The helpers an agent can call on

You do not assign these by hand. The agent reaches for the right sub-agents based on what the task needs. Here are the patterns you will see most.

Multi-perspective review

Legal, Technical, Sales, and Design sub-agents check a draft at the same time, each from its own angle. Cross-functional flags without four separate meetings.

Deep research

A question splits across sources, each sub-agent digs into one, and the findings come back synthesized instead of as thirty open tabs.

Ambiguity probe

Each requirement gets examined on its own for vague language, undefined metrics, and missing scope, so gaps surface before engineering does.

Edge-case hunt

Sub-agents pressure-test the work, looking for what breaks: the empty state, the abusive input, the path nobody specified.

Option explorer

Several approaches get drafted and compared in parallel, so you weigh real alternatives instead of the first idea that landed.

Cross-check

Claims get verified against your shared context and connected data, so the output holds up to your own product reality.

How It Works

Spawn, work, merge

The whole cycle runs under one request. You see a single, thorough answer come back, with the work behind it open to inspection.

01

Spawn

The agent breaks the task into angles that can be handled on their own, then spins up a focused sub-agent for each one. No setup from you.

02

Work in parallel

The sub-agents run at the same time, each with a single job and your shared context in hand. Four perspectives take about as long as one.

03

Merge

Their findings come back together as one answer, with each flag traceable to the sub-agent that raised it. You see the conclusion and the reasoning.

Frequently Asked Questions

Questions about sub-agents

A sub-agent is a focused helper that a main agent spins up to handle one piece of its task. Instead of a single model making one pass at everything, the agent delegates parts of the work to specialized sub-agents that run in parallel, then merges what they find into one answer.
Get Started

Get a team's worth of thoroughness

Let your agents bring their specialists to every hard task. 14 days free, cancel anytime.

Explore The Platform

How sub-agents fit in

Platform

Multi-agent orchestration

Sub-agents add depth inside one agent. Orchestration coordinates several agents across a whole workflow toward a goal.

Learn more →
Platform

Shared context

Every sub-agent works from the same product context, so each angle is grounded in your real product, not a guess.

Learn more →
Platform

Memory

Sub-agents apply what your agents have learned, so a review reflects the decisions and preferences you have already set.

Learn more →
Agent

AI PRD Writer

Sub-agents in action: its multi-perspective review checks every PRD from four angles at once before your team sees it.

Learn more →