Few people have done more to automate the office than Daniel Dines. So it is striking that the UiPath founder’s message on AI and jobs is a plea for patience, and a confession that he feels the anxiety too.
Dines built UiPath into one of Europe’s biggest software success stories by selling robots that do the repetitive parts of white-collar work. The company has since pushed hard into AI agents, most recently by buying the compliance-automation firm WorkFusion. Yet on the company’s podcast, The Path Forward, Dines spent much of his time warning against the very thing his tools enable: cutting staff in a hurry.
"Everybody feels some sort of anxiety, me included," he said, in conversation with UiPath colleague Andrada Morar. "We don’t know how our kids’ career is gonna look like." His answer to that unease is a line he repeats often. In times of anxiety, action is the answer.
No Einstein in the data centre
Dines is impatient with the biggest promise of the moment. Some in the industry talk of "50 million Einsteins in the data centre." He thinks that is only half right. A model, he argues, is an average of everything it has read. "An average by definition doesn’t have a taste."
He tested this himself, asking models to write fiction in a given style. The results came back bland. Taste, he says, comes from lived experience, not memory. He reaches for skiing to make the point. You can memorise every book ever written about the sport. It will not make you a skier. You have to fall on the slope.
That gap matters inside a company. Every enterprise runs the same handful of frontier models, with the same weights. Feeding them different data does not make them grasp your customer or your process. "Our memory is not our identity," he said.
Two ledgers, not one
His warning to executives is blunt. Do not read a job as a single output. Take a lawyer who reviews contracts. The visible outcome is a signed deal, and AI can speed that up. The hidden outcomes are harder to see. The same lawyer might mentor juniors, hold a client relationship together, or carry years of unwritten knowledge.
Dines wants firms to keep two ledgers, one for visible outcomes and one for hidden ones. Cut blindly, he says, and you destroy value you never measured. It is a pointed message from a man who sells automation. It also lands against a backdrop of real cuts. Carmakers have shed more than 20,000 white-collar jobs, and a growing chorus of bosses now pitch AI as a way to do more with fewer people. That is a sharp reversal from two years ago.
He also thinks the shift is slower than the hype suggests. Agents cannot simply plug into messy processes. Most firms have never mapped who is allowed to approve an invoice, or pay one. That knowledge sits in people’s heads and across departments. Documenting it will take years, he says, not a weekend.
The identity problem
The deepest worry in the conversation is not about tasks. It is about identity. Dines traces his interest in the subject to a lawyer friend. She told him her fear was not that her job would vanish. It was that her identity would become irrelevant. Many people build a sense of self around their work. He calls protecting that a shared human interest, and frames the human cost as the thing enterprises risk losing.
He is unconvinced AI will grow a self of its own. To him it is a tool, closer to electricity than to a colleague. He borrows an idea from an American philosopher of the 1970s, an argument that echoes Harry Frankfurt. There are two orders of will. A model can want something. Only a person can want to want something, to want to become better. Chasing a machine that truly reasons, he adds, would mean finding a way to inject pain, and risk building a Frankenstein no one understands.
Curiosity over credentials
Morar picked up the human thread. Models have memory, she said, but they lack the motivation to be excellent. AI can hand you knowledge. It cannot hand you curiosity, or the grit to push through when something breaks. She looks for those traits in her own team. She also argues that companies must still hire and mentor junior staff. Skip that, and there are no senior leaders in a few years.
There is a customer angle too. So much support has moved to bots that people now jab at their phones asking for a human. That friction, she suggests, is a clue about what only people offer.
None of this is disinterested. UiPath sells the agents and robots that make the cuts possible. A message that transformation is long, careful, and human-heavy also happens to describe a long, expensive engagement. Even so, coming from an automation billionaire, the caution is worth hearing. Governments are already counting the jobs AI touches. Dines’s bet is that the roles left standing will be richer, not poorer. The anxiety, his own included, is the price of not yet knowing.
To understand Dines’ perspective, it helps to know his background. Born in Romania, he moved to the United States in the early 2000s. He worked as a software engineer at Microsoft before founding UiPath in 2005, initially as a company focused on workflow automation. The firm pivoted to robotic process automation (RPA) around 2011, riding a wave of interest in automating repetitive office tasks. UiPath went public on the New York Stock Exchange in 2021, and Dines has since amassed a net worth of several billion dollars. His journey from a coder in a Bucharest basement to a Silicon Valley-style titan gives him a unique vantage point on the intersection of technology and human labor.
The broader context of Dines’ warning is crucial. Since the release of GPT-3 in 2020, and especially after ChatGPT’s launch in late 2022, the conversation around AI and employment has intensified. Many studies predict that large language models could automate up to 80% of white-collar tasks, from data entry to legal drafting. Yet the actual rollout has been uneven. Early adopters like car manufacturers and consulting firms have reported mixed results: while AI speeds up certain processes, it also introduces errors, bias, and a need for heavy oversight. Dines’ call for a dual-ledger approach addresses this gap—companies must weigh the efficiency gains against the loss of tacit knowledge that only experienced humans possess.
In an era where AI agents are being marketed as “co-pilots” for every role, Dines’ skepticism about taste and identity is a refreshing counterpoint. He echoes the philosopher Harry Frankfurt’s concept of “second-order desires”—the ability to reflect on one’s own wants and choose to become a different kind of person. Machines, no matter how advanced, lack this capacity. They can optimize for a goal, but they cannot question whether the goal itself is worthy. This philosophical foundation grounds his business advice: do not automate away the very things that make work meaningful.
For enterprises, the practical implications are significant. Dines suggests that the next five to ten years will be a period of slow but steady integration, not a revolution. Companies should invest in mapping workflows, preserving institutional memory, and retraining workers rather than terminating them. He advocates for reskilling programs that leverage AI as a teacher, helping employees transition from routine tasks to higher-value judgment calls. This resembles the approach taken by some Japanese manufacturers during earlier automation waves, where lifelong employment was protected by slow, deliberate adoption of robotics.
The takeaway for executives is clear: resist the pressure to cut costs quickly. Instead, use AI to augment human capabilities, not replace them. The real competitive advantage lies not in the cheapest operations, but in teams that combine machine efficiency with human creativity and empathy. Dines’ message, delivered from the heart of the automation industry, carries weight—precisely because he stands to gain from the very changes he is urging caution around. His humility in admitting his own anxiety about the future makes the argument even more compelling.