AI completes less than one task in ten in simulated business environments

In a simulated corporate experiment, programming AI agents completed less than one task in ten. This result particularly concerns Claude Opus 5.5 and GPT-6 Astra.
Anthropic Claims Massive Use of Claude
Anthropic asserts that Claude produces over 80% of its production code. In standard evaluations, these agents approach top performance; tests such as SWE-bench are nearing saturation.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
The Simulation Provides Colleagues for the Agents
The experiment assigned them roles as digital employees in a fictional software company, a scenario presented as a first. Their environment included colleagues as well as boards dedicated to projects and archives of conversations.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.