Challenges, Limitations, and Learnings

On the technical side:

- VS Code + Claude Code: Smooth management of 40+ Markdown files

- MCP and agents: Efficient documentary research, automated verification

- Vercel: Deployment to production in minutes

- Spotify: Podcast publication <10 minutes

- Short sprint organization: 5 intensive days = concentrated and coherent production

On the editorial side:

- Three-movement structure: Facilitates writing and consistency

- Single source, three outputs: Time optimization

- Verification agents: Zero chronological confusion detected

Image generation:

- Quality and control still inferior to text generation

- More variables, more errors

- Solution: Deliberately simple style, but needs improvement

Audio generation:

- No native integration in VS Code/Claude Code

- Workaround: ElevenLabs externally, but requires the studio interface for chaptering

- To explore: ElevenLabs API for future automation

Production capacity:

- Even with doubled token capacity (Anthropic plan), only 40% used

- Indication: The workflow is efficient, no need for maximum capacity

Technical stability:

- VS Code crashed only once in 5 days (token limit on Perplexity API)

- Very good overall stability

On reception:

- Will the editorial style with metaphors appeal or seem "too literary"?

- Is the long format (1,500 words) suitable for the web's short attention spans?

- Will the audience embrace a non-Western vision of AI?

On legitimacy:

- The author is neither a professional historian, nor a recognized essayist, nor a computer engineer

- Risk that the project remains "in the web archives" without visibility

Accepted bet: Relying on content quality, research rigor, and positioning originality to create its own legitimacy.