Skip to content
Recommendations In app: /agent-designUpdated 2026-11-15

Agents

What an agent needs to know before it can be built.

Agents lists the agents the plan needs and what each one has to know before it can be built.

Where: Recommendations, Solution, Agents. The page appears when the plan contains agents.

By default there is one agent per process. The list groups agents by area and shows, for each, how many of its 13 sections are ready, how many are still to fill, and what it saves.

The Agents list, grouped by area

The agent page ​

It opens with what the agent saves, the time per item before and after, model usage, and the build with its payback. Below are 13 sections in six groups:

  1. What it does: Starts when, Steps.
  2. Where a person is involved: Owner, Human review, Never without a person, When unsure.
  3. Systems and data: Systems and access, Data.
  4. How it decides: Decision rules, Messages.
  5. Success and testing: Success targets, Test cases.
  6. Runs on and build.

The Steps section shows how each step is done — by a rule, a single model call or the agent itself — with minutes, tokens and saving.

An agent page: the steps it covers and what is still to fill

The sections about people ​

These are filled in with the process owner: who owns the agent, where a person reviews, which actions never happen without a person, and who takes over when the agent is unsure. kawaru.app marks its own proposals "suggested".

Where a person is involved: review points and actions that always need a person

For each system the agent uses, Systems and access sets how it gets in: a platform connector, the system's API, a connector to build, or an MCP server.

Result: an agent with every section filled, ready for its spec. Show spec and Export turn it into a document for whoever builds it.

Why this step: an agent is built from these answers. A gap here becomes a question during the build or a mistake in the pilot.