Direction 04 of 05Agents that check each other
Agents that check each other.
One agent works, a second checks, a third looks for contradictions. Only what passes the checks reaches you. A person gives the approval.
Agents that check each other: Starting point, What changes, Result. Animation in three steps, the same content is available as text.
01 Starting pointA document passes through several hands. An error shows up late.
02 What changesOne agent drafts, a second checks against the requirements, a third looks specifically for contradictions. Whatever fails goes back.
03 ResultOnly what passes the checks reaches you. Every run leaves a report. A person gives the approval.
Illustration, simplified. Not a client case. On to the examples ↓
01 / 03Examples
What this could look like for you.
01
Draft, check, cross-check
- Today
- A document passes through several hands, mistakes surface late.
- After
- One agent drafts, a second checks against the requirements, a third looks specifically for contradictions.
- Stays with your people
- The approval.
02
A second opinion from another model
- Today
- A single AI model checks its own work.
- After
- A model from a second provider checks independently. Every finding is either backed up or dropped.
- Stays with your people
- The decision when they disagree.
03
A report instead of a black box
- Today
- It is unclear what an automation actually did.
- After
- Every run leaves a report: what was done, what was checked, what is still open.
- Stays with your people
- The questions the report raises.
Simplified examples, not client cases.
02 / 03Other directions
Or something else entirely.
Describe your task in your own words.
The entry step: outside view first, guide price CHF 2,90003 / 03Process
Four steps. One fixed price.
Only what is agreed beforehand gets built.