AI-Native Organisations Run on Skills: How to Structure and Scale Them — Imad Touil, QuantumBlack
AI Engineer World's Fair 2026 · 20:30
← All organizations
AI consulting and engineering
QuantumBlack
QuantumBlack, AI by McKinsey, helps organizations develop AI-enabled businesses and deploy machine-learning and agentic AI systems. Its work covers AI strategy, operating-model design, capability building and responsible deployment across customer, commercial, service and operational workflows. QuantumBlack Labs develops engineering tools and reusable technology; its offerings include Horizon, a set of AI development tools.
Founded in 2009, QuantumBlack began with work in Formula One before expanding its application of AI to business challenges. It is now led by Debasish Patnaik, a McKinsey senior partner who also helps shape its platforms, product suite and enterprise deployment approaches. Its engineering contributions include Kedro, an open-source Python framework that it donated to the Linux Foundation.
Kedro helps data engineers and data scientists turn experimental code into reproducible, maintainable pipelines. It separates data access from processing logic: a data catalog handles loading, saving and file-based versioning, while pipeline abstractions resolve dependencies between Python functions. Templates and reusable components support collaboration, and deployment options span single machines and distributed infrastructure. The framework is hosted by the LF AI & Data Foundation.
Explore the recordings
The supplied QuantumBlack archive contains one recording, offering several paths into Imad Touil’s proposed approach to structuring AI-native organizations. This guide describes claims and examples from that recording; it does not establish QuantumBlack’s current practices, product capabilities, or measured results.
Start with skills and the agentic stack
Watch AI-Native Organisations Run on Skills for Touil’s argument that reusable skills hold an organization’s executable know-how. Follow his distinctions between skills, hooks, sub-agents, and MCP servers, then his explanation of how they compose into workflows. This provides the conceptual foundation for the recording’s organizational recommendations.
Follow the path from individual skills to a shared platform
Return to Touil’s recording for the implementation and operating-model path: modular skill design, progressive disclosure, individual creation and testing, team collaboration, and a centralized catalog. His proposed platform includes discovery, dependency tracking, versioning, access control, evaluation, and observability. Treat these as recommendations presented in the talk, rather than verified features of a current QuantumBlack offering.
Examine governance and the limits of the evidence
Use the same talk to explore the regulatory review example and the risks Touil identifies: duplication, unclear ownership, quality decay, composition conflicts, prompt injection, embedded scripts, and sensitive business logic exposure. The example illustrates workflow composition; it does not establish compliance assurance. His illustration involving 15 teams over six months is a simulation, not measured organizational evidence. This path is useful for assessing the responsibilities and guardrails his proposal would require.
1 talk
Newest first1 speaker at AIE
Affiliations reflect their AIE appearances, not necessarily current employment.