# AI Engineer Paris 2026 Opening Keynotes: Mistral, Langfuse & Sizzy | Day 1

## Executive summary

The keynotes covered the profound economic impact of AI as a General Purpose Technology (GPT), detailing historical parallels with the Industrial Revolution. The technical discussion highlighted the evolution of AI development from simple prompting to complex, orchestrated 'software factories.' Speakers emphasized the shift toward robust, enterprise-grade solutions, focusing heavily on agentic safety, data sovereignty, and unified platforms that manage diverse tools and connectors.

## Key takeaways

- AI as a General Purpose Technology (GPT): AI is positioned as a GPT, similar to electricity or the internet, capable of driving long-term economic growth. However, realizing this growth requires more than just invention; it demands diffusion, investment, and organizational change (the 'slow clock' of adoption).
- The Shift to Software Factories: AI development is moving beyond simple 'vibe coding' and 'meat proxy' roles toward structured 'software factories.' This requires sophisticated orchestration to manage tasks, testing, and code generation in a controlled manner.
- Focus on Agentic Safety and Sovereignty: Enterprise AI requires robust controls to manage risks. Key concepts include dynamic privilege access, runtime guardrails, and data sovereignty (knowing where data is stored and processed) to build trust and compliance.
- Advanced Orchestration Tools: Modern tooling is moving toward unified, modular systems (e.g., using `Proxmox` or `Exeutor`) to manage multiple AI models and services from a single endpoint, ensuring portability and control.

## Technical details

- General Purpose Technologies (GPTs): GPTs are pervasive technologies (e.g., water, steam engine, electricity, computer) that improve across all sectors and raise the returns of their applications over time. AI is claimed to fit this definition.
- AI Development Paradigms: The process is evolving from simple prompting and 'meat proxy' roles to structured 'software factories' that automate tasks, testing, and deployment.
- Agentic Safety and Guardrails: Protecting against risks like prompt injection requires implementing runtime guardrails, defining classifiers (e.g., PIIs, secrets), and using policies to restrict agent privileges at the network level.
- Infrastructure and Sovereignty: Building a fully sovereign AI stack requires dedicated data centers (e.g., 10 MW facility near Paris) to control data storage, inference, and training, ensuring compliance and security.
- Orchestration and Sandboxing: Advanced workflows require sandboxing (e.g., using mini VMs via `Proxmox` or `Podman`) to ensure tasks are isolated and cannot compromise production environments. Tools like `Exeutor` and `Skillbox` aim to unify connectors and skills.

## Practical implications

- For build engineers, the focus must shift from merely integrating models to building robust, orchestrated systems (software factories) that manage state, testing, and security.
- Prioritize adopting sandboxed execution environments (mini VMs) for all agentic workloads to prevent unauthorized access or data leakage.
- Implement centralized orchestration layers (like `Exeutor`) to manage diverse model APIs and connectors, ensuring portability and reducing technical debt.
- Focus on implementing policy-based guardrails and classifiers to enforce enterprise security rules, rather than relying solely on model prompts.

## Topics

AI Engineering, General Purpose Technologies, Software Architecture, Agentic Workflows, Data Sovereignty, CI/CD, Mistral, Langfuse, STATION F, GPTs, Proxmox, Exeutor, Skillbox, Vibe, Anti-Slop

Source: https://www.youtube.com/watch?v=CGq9KRSb9Kc