Gemini 3.1 Pro contrôle mon PC en agent IA cowork ! Voici le résultat

Value of the Information & Strength of the Argument

The video offers practical value by demonstrating a concrete method to extend Gemini 3.1 Pro’s capabilities beyond a simple chat interface, addressing a real limitation (lack of persistent memory and file access). The argumentation is centered on the idea that traditional prompts are obsolete for agentic AI, and that structured prompts with hypotheses, scoring, and validation are necessary. However, the claims about benchmark scores (e.g., ARC-AGI 2, GDPval) are presented without specific sources or verification, and the demonstration relies on a third-party interface that is not clearly identified. The creator’s argument is persuasive for beginners but lacks technical depth and critical analysis of the limitations or potential risks of the approach.

Scientific Rigor, Source Quality, Title Accuracy

The video cites several benchmarks (ARC-AGI 2, GDPval, APEX Agent) and mentions competitor models (Claude Opus 4.6, GPT-5.2), but provides no direct links or references to these sources. The description includes links to the creator’s own training site and a tool link, but no official Google documentation or research papers. The title accurately reflects the content, which is a practical demonstration. The video includes a promotional segment for the creator’s training courses, which is not penalized but is noted. The overall scientific rigor is low due to the lack of verifiable sources and the promotional nature of the content.

The title accurately reflects the content: the video demonstrates how to configure Gemini 3.1 Pro to control a PC via an AI agent interface.

6/10

The video provides a practical demonstration of configuring Gemini 3.1 Pro as an AI agent with memory and file access, but relies heavily on unverified benchmark claims and promotional content for the creator's training courses. The technical explanations are superficial and lack rigorous sourcing.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction: converting Gemini 3.1 Pro into an AI agent to control the PC.
• 00:30Discussion of Gemini 3.1 Pro's architecture and benchmarks (ARC-AGI 2, GDPval).
• 02:00Explanation of the need for new prompting methods for agentic AI.
• 04:00Comparison of Gemini's lack of file access vs. ChatGPT's ability to write to virtual disk.
• 06:00Setup of the third-party interface, API key generation, and MCP configuration.
• 08:00Creation of a memory directory for Gemini 3.1 Pro.
• 10:00Demonstration of the AI creating and updating memory files.
• 12:00Explanation of prompt structure for large contexts and scoring prompts.
• 14:00Example task: web research on Gemini 3.1 with human-in-the-loop validation.
• 16:00AI generates the final report and saves it to the disk after user approval.

• Parlons IA Training Site — Creator's training courses for AI and business.
• SEO Agent IA Tool — Tool link mentioned in the description for testing.

• Google AI Studio — Official platform for accessing Gemini models and generating API keys.