---
sheet_format: 1.0
id: 2026-03-forkit-hanoi-sovereignty-ai
title: "Digital Sovereignty & AI: The Strategic Cost of Inaction"
speaker: Erwan Rougeux
with: []
date: 2026-03-07
event: "Fork It"
place: "ASA Coworking Space, Hanoi, Vietnam"
format: talk
language: en
duration: "15:27"
video: https://www.youtube.com/watch?v=ZVbVGuLUxcI
slides: https://slides.wanoo.me/2026-03-forkit-hanoi-sovereignty-ai/
topics: [digital sovereignty, ai agents, ai-assisted coding, geopolitics, vendor lock-in, open source, software economics]
source: "Summary written by Claude (AI) from the auto-generated YouTube captions, and the slides. Not a verbatim transcript; check the recording before quoting."
review: spot-checked
review_date: 2026-10-17
---

# Digital Sovereignty & AI: The Strategic Cost of Inaction

**In one sentence:** In a short strategic talk at Fork It Hanoi, Erwan Rougeux argues that waiting on AI is a competitive loss, but adopting it blindly costs you your digital sovereignty. The answer is to accelerate while keeping control over where and how agents run.

## TL;DR
Erwan opens with "deploy fast or die". The AI Pandora's box is already open, and whatever you decide, your competitors will use it. Waiting for the tools to settle lets them pull ahead. Using AI without control brings four risks: dependence, exposure, leakage and lock-in. He links these to geopolitics and to open-source licensing. His alternative is "parametric" coding: the human sets the intent and the guardrails (architecture, tools and libraries, security, UX, documentation, maintainability), and agents write and deploy the code in a continuous loop. Because developers spend less time on code, they have more time for control, jurisdiction, traceability and resilience. His thesis: the winners will be those who control how AI is integrated into production, not merely those who use it.

## Erwan's key points
- Erwan argues that refusing AI is not a neutral choice. Your competitors will use it anyway, and the time you lose waiting gives them a large advantage.
- He says AI has moved from lab experiments, in the early ChatGPT days, to code generated for production about three years later.
- He describes documentation, integration and testing as painful work that "generates code, not value", and says AI is well suited to it.
- He warns that using AI without control brings four risks: dependence, exposure, leakage and lock-in. AI access could one day be restricted by country, laws or geopolitical status, so dependence on Claude or OpenAI is a long-term risk.
- On exposure, he asks how your intellectual property is protected once your whole codebase is sent to Claude Code or Codex. He cites an article about an open-source project that AI rewrote under a new license with only 1.6% of the original code left. In his view, that kind of rewrite threatens open-source licensing, which underpins all cloud infrastructure.
- He says that since ChatGPT 3.5 he has not written a line of code or done a deployment by hand. He gives agents full control, but inside guardrails he defines.
- He compares this to parametric computer-aided design: the constraints together define the object, and here they define production-grade software.
- He describes a flywheel: speed creates scale, scale creates adaptation to users, adaptation creates leverage, and leverage increases speed.
- He argues that freed from code, the developer's job becomes making sure the application runs with the right control, jurisdiction, traceability and resilience. He cites drone strikes on AWS data centres during the war in the Middle East as an example.
- From his travels he observes that Southeast Asia is adopting AI far more readily than Europe.

## Outline
- [0:02](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=2s) Framing: a strategic and diplomatic talk rather than a technical deep dive. AI has turned digital sovereignty from a technical option into a strategic choice.
- [0:28](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=28s) "Deploy fast or die": if you are not first to market, you have already lost.
- [0:52](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=52s) Two camps, "full AI" and "no AI". Reality is not black and white, because the Pandora's box is already open.
- [1:39](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=99s) The cost of waiting for new tools, MCPs or model versions: competitors who have embraced AI coding move ahead.
- [2:29](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=149s) From AI as a lab toy after ChatGPT to AI-generated production code. AI belongs in every part of the business.
- [3:19](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=199s) Documentation, integration and testing as low-value work. Code is not value unless it is used.
- [3:42](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=222s) "AI is not enough": using AI shifts responsibility for writing code, and uncontrolled use creates deep problems.
- [4:08](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=248s) Risk 1, dependence: a scenario of AI tiers reserved for Americans, their allies and their rivals. "Who controls AI will control the next thing."
- [5:42](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=342s) Risk 2, exposure: intellectual property sent to Claude Code or Codex, and the anecdote about an AI-rewritten, relicensed open-source codebase.
- [7:15](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=435s) Risks 3 and 4: leakage of business knowledge and workflows, then lock-in once the AI decides how the software is built and run.
- [7:42](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=462s) No more humans writing code: he has done no manual coding or deployment since ChatGPT 3.5, but he wants guardrails.
- [8:50](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=530s) The parametric CAD analogy, and full agent autonomy (skipping permission prompts) bounded by constraints on architecture, provider, libraries, security, UX, documentation and maintainability.
- [10:02](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=602s) The developer's job is features that users use. With the right inputs, agents produce production-ready code. The intent, parameters, agents and code loop.
- [11:12](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=672s) New software economics: the speed, scale, adaptation and leverage flywheel.
- [11:59](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=719s) Without thought, you lose autonomy and sovereignty. The new job is control, jurisdiction, traceability and resilience, illustrated by the attacks on Middle East data centres.
- [13:14](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=794s) Thesis: the winners control how AI is integrated into production. Southeast Asia is embracing AI more than Europe.
- [14:22](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=862s) Closing: accelerate AI, keep control, build strategically. A late self-introduction as a European competitor to AWS, GCP and Azure, with a free-credit offer, then questions.

## Notable quotes
> "Deploy fast or die. That's the game." — Erwan Rougeux, [10:48](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=648s)

> "AI is the new industrial revolution. Who control AI will control the next thing." — Erwan Rougeux, [5:19](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=319s)

> "Our job is not creating line of code. Our job is create features that our user will use." — Erwan Rougeux, [10:02](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=602s)

> "The winner will not be those who use AI, but those who control AI and is integrate to production." — Erwan Rougeux, [13:14](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=794s)

## Questions this answers
**Q: What does Erwan Rougeux think about companies waiting before adopting AI?**
A: He thinks waiting is a strategic cost. The Pandora's box is open, and competitors who adopt AI coding gain ground while you wait for tools to mature ([1:39](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=99s)). His slides say the biggest risk is moving too slowly.

**Q: What are the risks of using AI coding tools without control?**
A: Erwan names four: dependence on a provider whose access could change with geopolitics, exposure of your code and intellectual property, leakage of business knowledge, and lock-in when the AI decides how your software is built and run ([4:08](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=248s), [7:15](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=435s)).

**Q: What is "parametric coding" with AI agents?**
A: By analogy with parametric CAD, the human defines the intent and constraints: architecture, cloud provider, libraries, security, UX, documentation and maintainability. Agents then generate and deploy the code in a feedback loop ([8:50](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=530s), [10:48](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=648s)).

**Q: Why does digital sovereignty matter more when AI agents write software?**
A: Erwan argues that once you stop thinking about code, your job is to make sure the application runs with the right control, jurisdiction, traceability and resilience. He cites strikes on data centres in the Middle East as an example of infrastructure risk ([11:59](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=719s)).

**Q: How does AI threaten open-source licences?**
A: He cites an article about an open-source codebase that AI rewrote and relicensed, with only 1.6% of the original code left, and which was legally treated as new software. He warns that this could hollow out the licences that all cloud infrastructure depends on ([6:04](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=364s)).

**Q: Does Erwan Rougeux write code by hand?**
A: He says that since ChatGPT 3.5 he has not written a line of code or done a deployment manually. Everything is done by AI, but within guardrails he sets ([8:05](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=485s)).

**Q: How does AI change the economics of software?**
A: He describes a flywheel: speed creates scale, scale allows adaptation to users, adaptation creates leverage, and leverage feeds speed again ([11:12](https://www.youtube.com/watch?v=ZVbVGuLUxcI&t=672s)).

## Mentioned
ChatGPT, ChatGPT 3.5, OpenAI, Claude, Claude Code, Anthropic, Codex, MCP, AWS, GCP, Microsoft Azure, MIT license, GPL, Creative Commons, computer-aided design (CAD), Middle East / Iran war, Southeast Asia, Europe, Fork It, Clever Cloud (implied, not named)

## Slides outline
1. Digital Sovereignty & AI: The Strategic Cost of Inaction — title. Notes: a strategic and geopolitical conversation, not a technical deep dive.
2. The biggest risk is not moving too fast, it is moving too slowly — waiting costs learning curves, productivity, competitiveness and optionality.
3. AI is becoming a core capability — productivity, operations, decision-making and software. The talk focuses on software.
4. Using AI is not enough — dependence, exposure, leakage, lock-in. "Accelerating without control creates faster dependency."
5. Building trust through orchestration — from generation to constrained orchestration: intent plus agents, bounded by architecture, tools and libraries, security, UX/UI, documentation and maintainability.
6. Parametric coding by AI agents — the intent, parameters, agents, code and feedback loop.
7. This changes the economics of software — the speed, scale, adaptation and leverage flywheel.
8. If agents build your systems, infrastructure matters even more — a controlled perimeter of control, jurisdiction, traceability and resilience around agents, code, data, secrets and workflows.
9. The winners will not be those who use AI, but those who control how AI is integrated into production — the core thesis.
10. Accelerate AI. Keep control. Build strategically. Thank you.