Gleb Tsipursky urges cross-functional AI

Generative AI strategy is moving out of the lab and into org design. That’s the real point of Gleb Tsipursky’s May 5 essay — not that teams should “collaborate” in some vague way, but that companies need formal cross-functional groups to decide where AI goes, how it gets used, and what risks it creates. His argument is simple: siloed pilots look fast at first, but they usually stall when they hit workflow reality, employee resistance, or governance problems. So the news here is less a technical breakthrough than a management claim about what AI deployment now requires. ### What is he actually arguing? Tsipursky says generative AI works best when IT, finance, HR, marketing, operations, and frontline staff all have a seat at the table. The point is not consensus for its own sake. The point is that each function sees a different failure mode — budget blowouts, compliance trouble, bad ideas in projects from selection through rollout. ### Why not just let IT handle it? Because most AI failures are not really model failures. They’re business-integration failures. An IT team can wire up a tool, but IT alone usually can’t define the right use case, judge whether outputs fit real operating conditions, or predict how workers will react when automation lands in their process. Tsipursky’s whole case rests on that gap — technical deployment is only one slice of adoption. ### What example does he use? He points to a mid-sized manufacturing company that wanted AI for demand forecasting and process automation. Leadership first assumed the IT department could manage the whole thing. But the project changed once marketing helped shape forecasting inputs, operations added ground-level processes, improving forecasting accuracy by over 30% and cutting waste by 25%. That’s the most concrete detail in the piece — and basically the proof point he wants readers to remember. ### Why does buy-in matter so much? Because employees resist tools that feel imposed by people who don’t understand their work. That’s especially true with generative AI, where the fear is not just “this won’t work,” but “this will change my job and nobody asked me first.” Cross-functional governance turns rollout into participation. It gives departments a chance to challenge assumptions early, which often lowers resistance later. ### Is this really about risk? Yes — more than the upbeat framing first suggests. When finance, HR, legal-minded operators, and business owners are involved early, companies are more likely to catch hidden data exposure, bad incentives, weak oversight, and use cases that drift away from strategy. In other words, the committee model is doing two jobs at once: finding value and containing damage. ### Why is this showing up now? Because 2024 and 2025 were full of pilots. By 2026, the harder question is which experiments become durable systems. That pushes companies toward governance structures — centers of excellence, internal portals, training programs, and cross-functional steering groups. Tsipursky has been urging companies to adopt these structures and companies start treating it as an operating model. ### Who benefits from this framing? Consultants, change-management teams, and internal AI governance leaders do. If the bottleneck is no longer model access but coordination, then the scarce skill is not just technical implementation. It’s getting different departments to agree on priorities, controls, and rollout plans without killing momentum. That makes “cross-functional AI” a services story as much as a tech story. ### Bottom line? Tsipursky’s piece is really a sign of where enterprise AI has gone. The winning question is no longer “Can this model do something impressive?” It’s “Can the organization absorb it without breaking trust, process, or accountability?” His answer is a committee. Maybe not a glamorous one — but probably a very 2026 one.