Tu n’utiliseras plus JAMAIS l’IA sur le web après cette vidéo

Item: Tu n’utiliseras plus JAMAIS l’IA sur le web après cette vidéo
Rating: 4
Author: PostSingularity Institute

You will NEVER use AI on the web again after this video

Title history

Tu n’utiliseras plus JAMAIS l’IA sur le web après cette vidéo — observed on 20261112-0057 UTC (YouTube)

by Thomas Berton · View the channel on PSI

Humanities, Social Sciences & Thought AI & Society UBDigital and Information technology: general topics UBJDigital and information technologies: social and ethical aspects

🎙 Thomas Berton 👥 48K 📅 December 10, 2025 ⏱ 42 min 👁 18K 📄 tutorial 🧭 2026-11-12

Available in: English (current) Français

Keywords

CLIagentcontextterminalworkflow

Summary

The video is a practical tutorial on using AI through command-line interfaces (CLI) instead of web browsers. The presenter argues that browser-based interactions are inefficient due to context loss, scattered chats, and constant copy-pasting. He introduces two main tools: Google Gemini CLI and Claude Code. For Gemini CLI, he explains installation via npm, authentication with a Google account, and basic usage. He highlights its ability to access local files, run shell commands, and create persistent context files (gemini.mmd) that allow resuming projects seamlessly. For Claude Code, he demonstrates similar features but emphasizes its agent system, which lets users delegate tasks to specialized sub-agents with independent context windows. He also shows how to create custom agents, manage permissions, and monitor token usage. The video includes a real-world example of scraping YouTube titles and generating an optimized title, illustrating a multi-step workflow. The tone is enthusiastic and productivity-focused, with a clear step-by-step approach. The presenter uses a French narration, and the content is aimed at intermediate users familiar with basic terminal commands. He also promotes his own community resources and services.

179 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides concrete, actionable information with clear demonstrations. The author presents several CLI features — such as persistent context files, parallel agents, and permission controls — that are genuinely valuable for improving productivity. The argumentation is logical: it contrasts the fragmented browser workflow with the centralized terminal approach, and backs it up with live examples. However, the claim of being ‘10x faster’ is not substantiated by measurements, relying mostly on anecdotal evidence. The demonstration of agent delegation in Claude Code is particularly compelling, showing how a task can be offloaded to a sub-agent with its own 200k token window, which is a significant advantage over conventional chat interfaces. The reasoning is consistent, and the step-by-step approach makes the techniques easy to replicate.

Scientific Rigor, Source Quality, Title Accuracy

The video does not cite scientific sources or official documentation within the content itself, but the description lists the official links for the tools (Gemini CLI, Claude Code, Codex CLI). The background and claims align with the features publicly documented on those official pages, lending a degree of credibility. The video is a tutorial rather than a research presentation, so the lack of academic citations is expected. The breadcrumb trail of links in the description helps viewers verify information. The title is sensationalized (’never again’) but the core content indeed suggests a shift from web interfaces to terminal-based tools, so the adequacy is moderate — the title reflects the enthusiasm of the author but overstates the absolute replacement. The audience’s reception (17.5k views, 737 likes) indicates positive reception, though no comment analysis is provided.

272 words

Title / Content Match

The title is catchy and slightly exaggerated, but the content delivers on its promise by showing how CLI tools replace browser-based AI workflows.

Quality & Reliability

8/10

The video demonstrates practical CLI AI tools with clear steps and genuine technical explanations. Claims about productivity gains are subjective and lack rigorous benchmarks, but the demonstration is accurate and reproducible.

Chapters

Cited Sources

Gemini CLI official site — Referenced as the tool for interacting with Gemini in the terminal.
Claude Code product page — Described as the author's preferred CLI tool with agent capabilities.
OpenAI Codex CLI quickstart — Mentioned as an alternative similar CLI tool.

External References

Contribution & Novelties

The video contributes a practical, non-developer-oriented introduction to AI command-line interfaces, emphasizing two features that are often overlooked: persistent context files that live on the local disk, and sub-agent delegation that creates isolated working sessions. It bridges the gap between developer-centric documentation and general productivity users, showing that these tools can handle writing, research, and file management without coding expertise. The demonstration of a persistent memory file (gemini.mmd / claude.md) is an original insight that helps users retain project context across sessions, and the agent system in Claude Code is presented with a clear use case.

Pour aller plus loin :

Command-line interface Wikipedia — Provides background on CLI evolution and its advantages.
Markdown Wikipedia — The context files are Markdown-based; knowing Markdown helps understand the structure.
ReAct: Synergizing Reasoning and Acting in Language Models — A research paper underlying agent behaviors, illustrating how LLMs can reason and act iteratively.

150 words

Radar Profile

The radar profile shows high scores in information quantity, information quality, and technical level, with a slightly lower reliability score due to the subjective nature of the productivity claims. This indicates a technically rich and well-structured tutorial that may lack empirical backing for its performance assertions.

Reliability 7/10