Stijn Christiaens

The through-line

Across the numbered material, Christiaens keeps circling back to one claim: organizations do not fail because of the technology itself, they fail because they repeat the same governance and data-quality mistakes with each new wave, and the current agentic AI wave is no exception. In his 2026 posts this becomes an explicit warning that governance for agentic AI must be structured before deployment rather than bolted on afterward . In the earlier interviews (2024-2025) the same instinct shows up as a more general observation: enterprises made the same errors moving to cloud data warehouses that they made on-premise, and the recurring obstacles are people and cross-functional collaboration, not the underlying tech ¹⁶. What has shifted is the object of concern — from data governance discipline broadly, to the specific infrastructure of "context" that agentic systems need in order not to fail .

On context as infrastructure

Christiaens treats context as the load-bearing layer beneath agentic AI, not a nice-to-have. He amplifies the framing that "Most agentic failures are failures of context" , and endorses the idea that "CDOs who can build and manage this infrastructure of context transform their businesses" . He is drawn to strategic arguments that context and meaning cannot simply be outsourced, flagging "Why You Cannot Rent the Meaning Layer" as good strategic thinking . He frames a well-run organization itself in these terms: "A well-governed company is a graph of loops running at different speeds" . He is also watching whether older tools for capturing context, like ontologies, have a real role in this cycle or are just a trend, asking "Are ontologies hot again, or a flash in the pan? Their fate is in the hands of your agent" , and separately musing on what "Unpacking context feels like" .

On governance before deployment

His clearest normative claim is that governance has to precede deployment: "enterprises aiming to take agentic AI to production need to structure governance before deployment. Retrofitting it after failure is far harder and more expensive than building it in from the start" . He connects this to a broader caution about mistaking motion for competence, highlighting that "the mistake is to confuse speed with judgment" . He also flags technical failure modes that reinforce the governance case, pointing to research on LLMs developing a "hive mind" syndrome when left to run unsupervised too long .

On data as an asset, not exhaust

Christiaens repeatedly argues organizations under-invest in the data layer and then relearn the lesson the hard way each technology cycle: "every new technology cycle organizations have to re-learn the importance of taking care of the data asset" . In interview, he names the two biggest misconceptions business leaders hold about AI: that it requires no data investment, and that AI systems are already reliable when they remain an emerging technology ¹⁷. He argues data should be treated as a valuable asset rather than exhaust, and that "AI ready data" spans both quality requirements for model training and the preparation of unstructured data, which he notes makes up 80-90% of organizational data, using semantic layers to turn documents into knowledge products ¹⁷.

From the stage

In the podcast and interviews, Christiaens goes further into the founder's psychology and craft than any written post does. He describes the discipline required to run a company for years: "you have to have strong opinions but loosely held," being stubborn enough to hold a thesis but open enough to change course "if the reality hits you in the face that you're going in the wrong direction" ¹⁹. He pushes back hard on romanticized entrepreneurship: "the romantic notion of the entrepreneur is a little bit of an illusion, but I do recommend start," while warning that "entrepreneurship is for everyone" is false and "the first years are typically very hard and very uncertain" ¹⁹. He details how Collibra's own category (data governance) was not the founding plan but was spotted during a pivot around 2009-2010 from semantic data integration ¹⁹, and lays out concrete, stage-dependent engineering metrics, investor-selection advice (choose the individual partner, not the firm brand, since that relationship can run 13+ years), and a sales philosophy that listening beats "demo barfing" all the features ¹⁹. He also offers the vivid time-compression image that "a day can feel like a week and a week can feel like a month" in a startup ¹⁹, and the observation that adoption speed is bounded by human and organizational adjustment, not technology, since "the future is already here, it's just unevenly distributed" ¹⁹.

Takeaways

• Build agentic AI governance before production deployment; retrofitting after failure is costlier than designing it in from the start .
• Treat "context" (semantic layers, ontologies, meaning) as core infrastructure for agentic systems, since most agentic failures trace back to context failures, not model failures .
• Don't confuse speed with judgment when evaluating AI-driven decision-making or agent autonomy .
• Assume unstructured data, roughly 80-90% of what organizations hold, needs deliberate preparation via semantic layers before AI can use it reliably ¹⁷.
• Expect to relearn data-asset discipline with every new technology cycle; the people and cross-functional collaboration problems recur more than the technical ones ¹⁶.
• When founding or joining an early-stage company, pick investors by the individual partner and align explicitly with co-founders on the scale of ambition, since misalignment on that scope causes the deepest conflict ¹⁹.