Roel Verbeeck

Roel Verbeeck's public statements cluster around two connected convictions: that meaningful digital products deserve their own identity, and that artificial intelligence today is a narrow, practical tool rather than the sweeping disruption often promised. Both threads run through his account of building Dokapi Public and his conversations about AI agents, and together they sketch a builder who is skeptical of hype but confident in concrete, workflow-embedded execution.

The clearest statement of his product philosophy comes from the Dokapi Public launch. When his team won the mandate from BOSA to renew the Mercurius platform into what he calls "het nieuwe digitale postkantoor voor de Belgische publieke sector," he was explicit that a solution of this scale could not be folded quietly into an existing portfolio. "Zo'n oplossing verdient geen bijrol in onze productportfolio," he writes, insisting instead that "Ze verdient haar eigen naam, haar eigen team, haar eigen roadmap" . This is a recurring instinct in how he talks about his work: important initiatives are not add-ons or line items, they are standalone commitments with their own resourcing and direction. The framing of the partnership itself, "Soms klopt een combinatie gewoon," suggests he sees the win less as a solo achievement and more as validation of the right team coming together for the right problem .

On AI, Verbeeck's tone in the "Virtual" podcast conversations with Davio Larnhout is deliberately grounded. Rather than speculating about general intelligence or distant futures, he anchors the discussion in what agents can actually do today: narrow, task-specific automation embedded inside real workflows ²³. His recurring point is that as these tools become widely available, the tools themselves stop being the differentiator. "Al die tools worden een commodity," he says, "en dus de mensen gaan een verschil maken vanuit hun eigen sterkte" ². This is a consistent worldview, not a throwaway line: as AI capability spreads, competitive advantage shifts back to human judgment, taste, and specific strengths rather than access to technology.

He backs this up with concrete client work rather than abstraction. Describing a project for Kan Design, he explains how Ixor decomposed a branding task into three cooperating agents: a digital twin representing the brand, a market agent that spins up large numbers of synthetic respondents segmented by age, gender, and belief, and a design agent that synthesizes both into proposals ³. This example matters to him because it addresses a specific, well-known failure in traditional research: people misreport their own preferences. He points to the Philips case where survey respondents chose colorful Bluetooth speakers when asked directly but took the black one home in practice, and argues that simulating market response with populations of agent personas can surface this gap more honestly than asking people outright ²³. It is a practical justification for agentic simulation, not a speculative one, rooted in a known bias rather than a promise of general capability.

Alongside this optimism about specific applications, Verbeeck is candid about disruption already underway. He notes plainly that fewer developers are needed to do the same amount of work in Belgium today, treating this as an observed fact rather than a future risk ³. He extends this logic to knowledge work more broadly, flagging sectors like accounting and Big Four-style consulting as likely early targets for the same kind of compression ²³. He does not frame this as alarmist; it fits his broader thesis that as tools commoditize, both jobs and value shift, and the people who adapt are those who lean into distinct human strengths rather than compete on tool access ².

Where he pushes back on overstatement is in his assessment of how far agentic AI has actually progressed. He uses the AlphaGo development arc as his benchmark, arguing that agentic AI has not yet had its "Move 37" moment, the point where the system produces a genuinely novel strategy that surprises and teaches human experts. Instead, he places current systems at the earlier "imitation stage," comparable to AlphaGo still learning by imitating grandmasters, while acknowledging that reasoning models are pointing toward a future where AI could eventually produce original strategies humans learn from ³. This is a carefully calibrated position: real progress, real disruption already visible, but a clear-eyed refusal to claim that agents have crossed into original strategic thinking yet.

Taken together, these sources present a consistent operator's mindset. Verbeeck moves fluidly between building tangible public-sector infrastructure and prototyping agentic systems for private clients, but his language stays anchored in specifics: named products with dedicated roadmaps, named client projects with measurable structure, named cognitive biases being addressed. He resists sweeping AI narratives in favor of testable claims about what agents solve today, while still tracking the trajectory toward more capable systems. The throughline is a preference for concrete proof over promise, whether that proof is a won government mandate or a market-research agent built to catch what people actually do rather than what they say ²³.