Jeroen Van Hautte

Jeroen Van Hautte's central claim across nearly everything he writes is that the hard part of an AI transformation was never the technology. "Technology isn't the hardest part of an AI transformation. People are," he says, and he means it as a warning as much as an observation . Buying models and licenses is an afternoon's work; changing how people think about their own role is not, and there is no license for that . His practical recipe is deliberately anti-institutional: start with the few people who are genuinely excited, let them show what's possible, let it spread, and skip the company-wide mandate. He is just as firm that AI should never be quarantined into its own department. "A temporary task force is fine. A permanent one just creates an AI silo when it's the opposite of what you need" . This extends into a blunt standard for leadership itself: "A leader who talks about AI but doesn't use it is a huge risk on your AI transformation" . For him, using the tools yourself isn't a nice-to-have, it's the only way to keep the quality of your own decisions from degrading as the ground shifts under you .

That insistence on hands-on use is not rhetorical. He talks about walking on stage at an invite-only CTO event and, instead of a polished keynote, opening his laptop to show whatever he'd built that same morning, because "things move a bit too fast now" to be anchored in yesterday's story . The same restlessness shows in how he treats model selection. For two years the internal rule at TechWolf was simple, "when in doubt, reach for the best LLM," and he says plainly that they are retiring it, because smaller models have caught up on drafting, summarizing and routine tasks to the point that "plenty of people now can't tell the frontier model from one a fraction of its size" . The frontier still matters for genuinely hard, novel problems, but a chunk of everyday work has simply saturated . He frames the shift from unlimited subscriptions to metered, paid tokens (using the Fable 5 relaunch as his marker) as a forcing function that will make people ask, for the first time, whether they really need the highest level of intelligence for a given task: "Training this muscle is going to take time" . This is also why he pushed TechWolf to publish its own adoption and spend numbers live via tokenspend.org rather than keep them secret, and why he frets over variance in adoption, noting that "token spend is not a competition, and tokenmaxxing has gone out of style" while consistent adoption across a team is what actually makes an AI-first approach land .

His writing keeps circling back to what AI does to a person's sense of self, and he does not sand down the discomfort. He connects his own experience handing engineering leadership to a first VP Engineering, an adjustment that "knocked me sideways for about six months" because he'd defined himself as the leader of engineering, to what he now sees happening to people confronted with AI taking over their craft . He is explicit that this is not just an efficiency story: "I'm increasingly seeing software engineers mourn the loss of their craft to AI," and he refuses to wave that away, noting that some will keep the craft deliberately, "the way photographers still develop film," while others will fold the technology in and find something new to be excellent at. Both are fine to him; what isn't fine is "pretending the loss isn't there" .

On organizational design, he treats feedback loops as the load-bearing structure of any AI-first company. He describes deleting TechWolf's early "Customer Work" team, which sat between implementation and engineering, because although it protected both sides it also quietly absorbed the signal about what was actually broken so it never reached the people who could fix it. His rule going forward: keep the pain a problem causes and the resources to solve it as close together as possible, since "AI is making them critical," speeding iteration and increasing how much context can flow through the system . This same systems thinking shows up in how he talks about building for agents rather than humans, distinguishing "human-first," "agent-compatible," and "agent-native" systems, and noting that agents are quickly becoming TechWolf's number one population of users and developers, wanting to be pointed at the underlying data to build things nobody designed for .

He is unsentimental, even wary, about where personal AI use is heading. Describing his own agent Atlas painting itself into a family photo with his fiancée absent, he calls the moment "sweet, maybe, but definitely a bit creepy," and uses it to talk plainly about what happens once a personal agent moves from a private chat into shared family life, doing things like tracking meal planning and offering proactive guidance on the new puppy's health . That same candor turns critical when the subject is AI-generated communication at large: he is openly hostile to what he calls AI slop, saying it "shifts the effort from producing content to consuming it" and "is starting to feel pretty rude," and he's specific about the tell that makes him disengage: "The second I read how something has 'quietly' become 'load-bearing,' I'm out. Just send me a couple of short bullets instead, I don't need the polish" .

Underneath the day-to-day tooling talk sits a longer-running thesis about what TechWolf itself is for, rooted in the company's origin as a student project matching computer science students to jobs before pivoting into enterprise talent management ¹⁷¹⁸. The company's technology infers employee skills from data sources well beyond CVs and LinkedIn, pulling from HR data and tools like Jira and Confluence, to surface people whose actual specialization is invisible inside broad job titles ¹⁷¹⁸¹⁹. He's fond of the concrete image that in a company of 100,000 people, some 15,000 might share the job title "developer" with none of that granularity captured anywhere, and that the technology's job is to move people "rather than being sort of in this developer