November 18, 2026
Host
Welcome to the podcast. Today we're exploring a strategic reflection on how Ubisoft can transform its knowledge management into an AI-ready production intelligence ecosystem. I'm joined by a guest who has been thinking deeply about this. Thanks for being here.
Guest
Thanks for having me. It's a topic I care about deeply. Ubisoft has accumulated an enormous amount of production knowledge over the decades. But much of it remains fragmented across projects, engines, and studios. And I believe generative AI and the Core Services transformation give us a unique chance to change that.
Host
So what's the core problem you're trying to solve? Is it just about better search?
Guest
Not just search. It's about moving from documenting knowledge for future consultation to activating it continuously during production. Today we capture a lot after the fact—post-mortems, closing kits, lore bibles. But teams often struggle to find and reuse that knowledge in their daily work. The challenge is fragmentation and lack of connection.
Host
You mentioned five strategic pillars. Let's start with the first: One Ubisoft Knowledge Ecosystem. What does that mean in practice?
Guest
It means moving from fragmented repositories to a unified access and discovery layer. Today knowledge lives in Confluence, SharePoint, engine docs, project repos, asset libraries, and more. Anvil has its own ecosystem, Snowdrop has its own. Each makes sense locally, but collectively it's a maze. A developer shouldn't need to know which studio or engine team documented a solution.
Host
But isn't some fragmentation inevitable? Different engines and studios have different needs. Forcing everything into one system could be counterproductive.
Guest
Absolutely. I'm not advocating for a monolithic platform. A federated approach makes sense for many technical systems—keep information close to its source. But we need to avoid unnecessary fragmentation. The first step is a clear inventory and common index of existing sources. Then identify where repositories overlap or duplicate. Finally, a selective consolidation roadmap based on usage and value.
Host
So it's about connecting the dots rather than merging everything. What about metadata and taxonomy? That sounds like a lot of work.
Guest
It can be, but we should keep it pragmatic. Metadata should help users find relevant info, understand which version is reliable, identify the owner, and know if they can access it. For example, tagging by game, engine, discipline, project, or confidentiality level—but only where it genuinely improves discovery. The goal isn't a perfect classification system; it's making knowledge visible and accessible.
Host
Let's move to pillar two: Knowledge by Design. How do you capture knowledge as it's created rather than after the fact?
Guest
Today a lot of KM relies on manual documentation, standards, and post-mortems. Those remain valuable, but we can shift toward capturing knowledge as part of production workflows. For example, AI can generate technical documentation from source code, capture decisions during production, structure information from approved meeting transcripts, or extract knowledge from project tools. We should also explore documentation-as-code and Markdown-based approaches.
Host
And AI could help maintain that knowledge over time—flagging outdated docs or contradictions?
Guest
Exactly. AI can identify outdated documentation, contradictions between sources, duplicated information, or missing content. But expert validation and human-in-the-loop mechanisms remain essential. The potential gain is significant: less manual effort, better relevance, and higher reliability.
Host
You also talk about integrating knowledge into production workflows. Can you give a concrete example?
Guest
Sure. When a project prepares a production gate, Core Services could provide an AI-assisted briefing that surfaces relevant lessons learned, risks, and recommendations from comparable projects. The team reviews and challenges those insights. Or when a team starts a new feature, they could retrieve existing standards, technical approaches, or reusable components through the tools they already use. We should start with a few pilots with volunteer productions.
Host
Pillar three is about live access to reusable code, assets, and production capabilities. How does that differ from what Closing Kits do today?
Guest
Closing Kits preserve source code, assets, and dependencies after a game ships—essential for long-term preservation and remasters. But the next opportunity is to make relevant code and assets available much earlier, while productions are still active. Imagine a team starting a new project and discovering existing gameplay systems or technical solutions from other Ubisoft projects. They could understand dependencies and decide whether to reuse.
Host
That sounds great in theory, but confidentiality and IP are huge hurdles. Not every asset should be accessible to everyone.
Guest
Absolutely, and we need to address security, legal, and confidentiality from the beginning. But we should aim for reuse by design, not rely on manual discovery. We can explore secure connectors to Perforce, Git, Anvil, Snowdrop, Unreal, and other pipelines. AI can help with asset classification, tagging, dependency analysis, and code understanding. Closing Kits would remain critical for preservation but become part of a broader ecosystem for continuous reuse.
Host
Pillar four: AI-native access and production enablement. How do you turn this collective knowledge into a foundational capability for generative AI?
Guest
Today AI assistants can write code, summarize documents, and automate workflows. But their value increases dramatically when they can securely access relevant, reliable, up-to-date Ubisoft knowledge. We should establish a shared enterprise knowledge-retrieval capability that internal AI assistants and approved agents can use. This includes permission-aware access to documentation, code, metadata, assets, and production knowledge, with semantic search and common APIs, including standards like MCP where appropriate.
Host
What would that look like for a programmer or a technical director?
Guest
A programmer could ask whether an existing gameplay feature has already been implemented on another project. A technical director could retrieve documentation on an engine architecture or compare approaches across productions. A team working on a legacy game could use AI to understand existing source code and dependencies. Or a small prototyping team could identify reusable systems and assets to accelerate their first playable. The key is that AI would be augmented by Ubisoft's own knowledge and experience—a significant competitive advantage.
Host
Pillar five is governance, ownership, and knowledge operations. Why is this the most important pillar?
Guest
Because technology alone won't solve fragmentation. We need clear ownership, common standards, and decision-making across Core Services. Today knowledge initiatives are managed independently by teams, engines, and services. Each defines its own approach to documentation, tools, taxonomies, and maintenance. If we want a coherent ecosystem, we need a shared mandate for knowledge strategy, architecture, taxonomy, and discoverability. We also need a network of knowledge owners and subject-matter experts across engines, projects, and Creative Houses.
Host
But doesn't centralizing governance risk slowing down teams and creating bureaucracy? How do you balance that?
Guest
The objective is not to centralize ownership of every piece of knowledge. It's to centralize the framework that makes decentralized knowledge discoverable, reliable, secure, and reusable. We need to clarify who is responsible for creating, validating, maintaining, archiving, and retiring content. Confidentiality, IP, access rights, and retention policies must be managed consistently. And with AI generating more content, curation becomes even more critical. We can't just accumulate information; we need mechanisms to ensure it remains useful and trusted.
Host
You also suggest that the Knowledge Management operating model needs to evolve. What does that mean for KM specialists?
Guest
Today our specialists focus on producing documentation, collecting information, formalizing best practices, and preserving knowledge. Tomorrow their role could increasingly involve knowledge architecture, curation, integration, governance, and AI enablement. They would become knowledge architects, curators, facilitators, and AI enablement partners. They'd work closely with technical and production experts, who remain accountable for the accuracy of their domain knowledge. We also see an opportunity to connect KM with Expertise Development and Learning activities, so curated knowledge feeds academies, mentoring, and expertise pathways.
Host
That sounds like a significant organizational shift. Who should lead this?
Guest
I believe our Production Expertise Development, Knowledge Management, and Game Preservation organization could lead the common knowledge strategy, taxonomy, content governance, curation model, and KM contributor network. Tech Services, IT, and the AI Hub would jointly define and deliver the technical architecture, platform capabilities, secure connectors, and AI integrations. Creative Houses and production teams would continue to own their domain knowledge and validate critical content. Global Learning and Expertise Development would help translate curated knowledge into structured learning. A cross-functional steering structure would align priorities and arbitrate decisions.
Host
Let's talk technology strategy. You mentioned five architectural layers. Can you walk us through them?
Guest
Sure. The first layer is our existing source systems: Confluence, SharePoint, Perforce, Git, engine documentation, production tools, asset catalogues, Closing Kits, and other repositories. The second is an integration layer with secure connectors, APIs, metadata extraction, and synchronization. The third is a knowledge layer with shared taxonomy, semantic indexing, relationships between assets and systems, content ownership, versioning, and permission controls. The fourth is an AI access layer enabling permission-aware retrieval and reusable services for internal assistants, agents, and production tools. And the fifth is the user experience: a unified discovery capability integrated into the tools people already use.
Host
That's ambitious. How do you avoid launching a massive transformation that stalls?
Guest
I recommend a progressive roadmap starting with a few concrete pilots. In the first three months, align on the mandate and sponsors, map existing repositories and initiatives, identify key stakeholders, and select the most relevant production use cases. Deliverables would be a strategic charter, initial inventory, target architecture principles, a governance proposal, and a selection of pilots. Between months three and six, launch two or three pilots. Between six and twelve months, expand the most successful pilots across additional projects, engines, and knowledge domains. Formalize the knowledge ownership model and contributor network. Over twelve to twenty-four months, progressively expand production integration, code and asset reuse, and AI-enabled knowledge workflows.
Host
How do you measure success? Is it the amount of documentation indexed?
Guest
No, I don't believe volume of documentation is the right measure. Success should be measured by impact on productions. For example: how much time does it take to find relevant documentation, expertise, code, or assets? How often are existing production capabilities reused across projects? Are we reducing duplicate work? How reliable and up-to-date is our critical knowledge? Are teams adopting the common discovery and AI-assisted workflows? How much time are we saving on documentation and knowledge maintenance? Ultimately, are these capabilities contributing to faster iteration, improved production efficiency, or reduced costs? We need baselines and targets from our pilots, but the focus must remain on measurable production value.
Host
What do you need from leadership to move forward?
Guest
Alignment on four key questions. First, strategic ambition: do we agree that Core Services should take responsibility for establishing a unified, AI-ready knowledge ecosystem across Ubisoft? Second, mandate and governance: can Production Expertise Development, Knowledge Management, and Game Preservation be empowered to lead the common knowledge strategy and standards, in partnership with technology and production owners? Third, technology sponsorship: can we establish a joint initiative with Tech Services, IT, and the AI Hub to define the target architecture and prevent further fragmentation? And fourth, resources and execution: can we dedicate a small cross-functional team to establish the roadmap, deliver initial pilots, and demonstrate measurable business value? We need to agree on direction, establish clear responsibilities, and give ourselves the means to start demonstrating what this could look like.
Host
You close with a conviction about activating what Ubisoft knows. Can you summarize that?
Guest
Ubisoft's competitive advantage is not simply the amount of knowledge, code, and assets we've accumulated. It's our ability to make that collective experience usable by every new production. Generative AI gives us an opportunity to turn accumulated knowledge into an active capability for game development—helping teams find the right information, reuse existing solutions, make better decisions, and focus more energy on creating distinctive games. The Core Services transformation gives us an organizational opportunity to make this happen. We already have many building blocks: Knowledge Management, game engines, technical experts, development pipelines, Closing Kits, Asset Management, Communities of Practice, and Expertise Development. What we need now is a shared vision, a clear mandate, strong governance, and the right technology partnerships to connect them. Ultimately, our ambition should be to move from documenting what Ubisoft knows to activating what Ubisoft knows. And I believe this is a strategic opportunity worth pursuing together.
Host
Thank you for sharing this vision. It's a compelling call to action for Ubisoft's knowledge strategy. I'm sure our listeners will find it thought-provoking. Thanks for joining us.
Guest
Thank you for having me. I hope it sparks some productive conversations.