Beyond the “Double Exponential”

Takeaway 1

The LAMP Stack is Dead, Long Live the “CLAW”

For twenty years, the LAMP stack (Linux, Apache, MySQL, PHP/Python) provided the architectural foundation of the internet. That foundation has been superseded by the CLAW stack: Compute, LLMs, Agents, and Workflow.

While the “open weights” movement has made significant strides, the forum participants warned that weights alone are insufficient for true sovereignty. Sovereignty is a hollow victory if a model only speaks the dialect of the latest proprietary silicon. For instance, a flagship open-weights model released just days before the forum required specific precision found only on “Blackwell-class” chips, effectively stranding the world’s existing “Hopper-class” inference capacity. It took a dedicated open-stack team three days to port that model across ten different chipsets using a cross-vendor substrate — a feat that highlights why we must defend the Seven Pillars of Open:

• Open Science — Published methods and reproducible results.
• Open Data — Transparent training corpora and provenance.
• Open Standards — Interoperable, royalty-free protocols.
• Open Source — Code under recognized licenses.
• Open Weights — Permissive model parameters.
• Open Platform — Tooling and orchestration available to all.
• Open Hardware — Specifications for silicon and accelerators.

“Open weights without an open compute substrate is a hollow victory.”

Takeaway 2

“Sovereign AI” is Just Branding Without Cryptography

As compute costs continue their precipitous decline, the strategic “moat” is shifting. The value no longer accrues to those who serve the cheapest token, but to those who own the verification layer.

The verifiability gap is widening: it is getting cheaper to run an AI, but increasingly expensive to prove what that AI actually did, what data it touched, and which chip processed the request. To bridge this gap, the forum proposed a three-tier model of verification:

1. Self-verifiable — The model justifies its own output.
2. Independently verifiable — A third-party observer confirms the action.
3. Cryptographically verifiable — Tamper-resistant, contemporaneous evidence captured at hardware boundaries.

“Sovereign AI without cryptographic verification is just branding.”

The most urgent technical requirement today is a deterministic control plane. We need binary-auditable evidence of an agent’s behavior — captured at the moment of action — to replace model-vendor promises with mathematical certainty.

Takeaway 3

The Junior Developer Pipeline is in Crisis

The job market is undergoing a radical bifurcation. While elite graduates command salaries of €150,000, the median graduate is struggling to find entry-level work. This isn’t because models are failing; it’s because our cognitive design is failing.

Cognitive design is the discipline of fitting algorithmic tools to the way humans actually work. Currently, roughly 19 out of 20 enterprise AI pilots fail because they create a Cognitive Black Hole — a state where an organization’s data is absorbed by an AI that never feeds insight back to the collective. The machine gets smarter, but the organization atrophies. To survive, education must shift from “syntax” to “specification.”

Takeaway 4

Friction is a Feature, Not a Bug

In the rush to make AI “frictionless,” we risk destroying the pedagogical basis of learning. The forum argued for friction-by-design — the intentional reintroduction of “productive struggle” into workflows.

When AI removes all effort, we encounter the calculator effect, where essential human skills — writing, reasoning, and social negotiation — atrophy. This is particularly dangerous in adolescence, where using AI as a “social substitute” during identity formation can displace the critical practice of human conflict and negotiation.

“Training and learning cannot be easy. It can be fun, but not easy.”

AI cannot yet replicate a teacher’s silent know-how — the tacit craft of judging a hypothesis or sensing when a proof “feels” right. We must design systems that protect the struggle, ensuring the machine doesn’t become a substitute for human growth.

Takeaway 5

Tokens are the New Digital Public Good

Compute, expressed as “tokens,” is now a strategic resource equivalent to the electric grid. However, we must distinguish between Free Tokens — marketing funnels that monetize user data — and Open Tokens which are governed, transparent, and provenance-verifiable.

The open-source ecosystem, currently “held together with dental floss and gum,” is under siege. AI-assisted code contributions have surged dramatically in just six months, burying human maintainers under an avalanche of unreviewed pull requests. To sustain this substrate, the Paris Initiative was established — a four-point commitment published on GitHub, covering:

1. Tokens — Establishing neutrally-governed token pools.
2. Compute — Providing “in-kind compute” to students and the Global South.
3. Governance — Creating standards for environmental and provenance disclosure.
4. Global Access — Ensuring the digital substrate remains a public resource, not a corporate monopoly.