Brief IA

Google Revolutionizes the Gemini API with Agent Skill Feature

💻 Code & Dev·Tom Levy·

Google Revolutionizes the Gemini API with Agent Skill Feature

Google Revolutionizes the Gemini API with Agent Skill Feature
Key Takeaways
1Google introduces the Agent Skill feature for the Gemini API, enhancing knowledge of updates and SDKs for AI models.
2The Gemini 3.1 Pro Preview model saw its success rate jump from 28.2% to 96.6% across 117 tasks thanks to this feature.
3Older models, such as version 2.5, benefit less from this feature due to more limited reasoning capabilities.
💡Why it mattersThis advancement allows developers to optimize the use of AI models by integrating up-to-date information, thereby enhancing the efficiency of coding agents.
Le brief IA que lisent les pros

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.

Choisis ton rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

📄
Full Analysis

Google Enhances the Gemini API with Agent Skill

Google has recently introduced a new feature called Agent Skill for its Gemini API. This innovation aims to address a major issue faced by AI-based coding assistants: their inability to incorporate updates and best practices after their initial training. With this skill, coding agents can now access up-to-date information on models, SDKs, and code examples.

In tests conducted on 117 tasks, the Gemini 3.1 Pro Preview model demonstrated a dramatic improvement in its success rate, rising from 28.2% to 96.6%. This skill, first introduced by Anthropic at the end of last year, has quickly been adopted by other companies specializing in artificial intelligence.

Impact on Gemini Models

The results of the Gemini models, both with and without the Agent Skill, show that the newer versions of the series 3 benefit significantly from this new feature. Google attributes this improvement to their more advanced reasoning capabilities. In contrast, older models, such as version 2.5, have shown more modest improvements, which Google explains by their more limited reasoning abilities.

Additionally, a study conducted by Vercel suggests that using AGENTS.md files to provide direct instructions to the models could prove even more effective. Google is also exploring other approaches, including MCP services. The Agent Skill is now available on GitHub, providing developers with direct access for integration and experimentation.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.