The method

Twenty years of creative innovation practice, encoded into AI agents that work together.

Method, techniques, tools, stances: our know-how is built into several agents, step by step.

Two stances that run through every phase

Dynamic stance

In two movements

01Diverge widely, without judging.

02Converge by sorting and selecting the strongest leads.

Purge · SCAMPER (seven ways to reframe) · direct analogies

Sensitive stance

In three movements

01Move away from the problem, toward the imaginary.

02Let things emerge in a blurrier, more symbolic space.

03Converge to make concrete what has emerged.

Imaginary journey · identification analogies

Without orchestration

Nearly 90%

of the tasks entrusted to teams of AI agents fail, for lack of orchestration.

It is not a matter of raw power. The same failures occur with GPT-4o or Claude. But when the agents are reorganized without changing the model, results climb by 9 to 16 points. The lever is orchestration.

The agents don't know what to do

Poorly described task, vague roles, repeated steps, an end that is never declared. The instruction is missing.

42% of failures

They each work on their own

One agent starts from scratch, another keeps information to itself, a third doesn't listen. Nobody holds the thread.

37%

Nobody checks the result

The system stops too early or delivers without rereading. Those with a verifier fail markedly less.

21%

In AI Innovation Box, each agent has a precise instruction at a precise step. All of them share the context built with you. And you arbitrate at every step: that is the verification the others lack.

Cemri et al., “Why Do Multi-Agent LLM Systems Fail?”, NeurIPS 2025, arXiv:2503.13657. Failure rates from 41% to 86.7% depending on the system, over 1,642 runs of 7 multi-agent systems.

The iasagora foundation

CPS (Creative Problem Solving), Design Thinking, C-K theory (Concept-Knowledge) and our two stances: a framework drawn from published research on creativity.

An active academic partnership

The IA . Innovation Club and the LaPEA (Université Paris Cité) put the tool to the test over time.

ia-innovation.club

La posture sensible, a book

The three movements of the sensitive techniques come from the LDS model (logical, dynamic, sensitive), described in La posture sensible (“The Sensitive Stance”), co-written by Guy Aznar and Stéphane Ely (Créa Université, 2010).

See the book

Need support to explore new categories, targets or user experiences? The iasagora consulting team is there for you.

Try it for free Talk about consulting support