Companies are about to give AI agents a lot more to do. According to UiPath founder and CEO Daniel Dines, that makes the human role more important, not less. Speaking at the company’s FUSION event in Dubai, he argued that agentic automation only works when people direct it and take responsibility for its actions, as first reported by The Next Web.
Dines spoke on Monday at the Museum of the Future. His message to customers was blunt: delegation to AI is inevitable, and the workforce needs to be ready for agentic automation before it arrives at scale.
Key Takeaways
- Daniel Dines emphasizes that the role of humans will become more critical as AI agents take on more tasks, requiring better training models.
- Dines advocates for hiring junior staff who learn from seniors to adapt to a world shaped by agentic automation.
- To implement agentic automation effectively, companies must name responsible individuals for AI actions and avoid using AI where conventional software suffices.
- The UAE government is a progressive partner in adopting agentic automation, with multiple sectors exploring advanced AI discussions.
- Leaders should treat agentic automation as a combination of staffing and governance, not just a technological upgrade.
Table of contents
An Education Model Built for the Factory Floor
Dines took aim at how people are trained. In his telling, the modern school system traces back to the industrial revolution, roughly 200 years ago, and was designed to turn out disciplined workers who carried out instructions from above. He said that approach won’t hold up in a world shaped by AI.
What he wants instead is staff who act on their own initiative. If agents handle a growing share of the routine work, the people around them have to decide what to ask for, check what comes back and step in when something looks off. That’s a different skill set from following a procedure.
He also expects companies to need fewer narrow specialists and more people who understand entire systems. His example was the legal profession. An AI agent can absorb the law from published sources, he said, but it can’t learn how that law plays out inside one specific company. That gap is where human judgment still earns its place in agentic automation.
Accountability Is the Limit That Doesn’t Go Away
Dines tied the Dubai talk to a framework he first laid out in Las Vegas in September, when UiPath launched Cartographer. He described three limits of AI: it doesn’t learn on the job, it isn’t exact, and it can’t be held accountable. The last one got the most attention in Dubai.
His reasoning was practical. Once a company lets an AI system act on its own, those actions carry consequences. Even a routine email from an agent to a customer can commit the business to something. Software can’t absorb the fallout, so a person has to.
For executives, this reframes a common question. Many deployment plans focus on what an agent is capable of doing. Dines pushed the conversation toward who signs off on what it does, and who answers when it gets something wrong. In agentic automation, naming that person early is part of the design, not an afterthought.
Don’t Use AI Where Plain Software Will Do
One of the more counterintuitive points came from the CEO of an AI-focused company. Dines told the audience that being able to use AI for a task is not a reason to do so. If ordinary software can handle the job reliably, that’s the better choice.
His argument rests on two properties. Conventional software can be tested, and it gives the same answer every time. AI outputs can vary, which is a problem for processes that need to be exact.
His suggested split is useful for anyone planning a rollout:
- Use AI to write the software for well-defined, repeatable work.
- Have people review and test that software before it runs.
- Keep agents for tasks where flexibility is actually needed.
Seen this way, agentic automation isn’t a replacement for every script and workflow. It’s the layer reserved for work that genuinely calls for judgment, sitting on top of tested software.
Dines said his own experience supports the approach. He wrote his book, The Work That Remains, with AI and rarely edited the text himself. From that, he concluded that people will stop touching code by hand. The human job shifts to reviewing and testing what the AI produces.
Keep Hiring Juniors, but Change How They Learn
Dines repeated his advice that companies should keep bringing in junior staff. There’s a catch, though. Juniors now need to work alongside senior colleagues from their first day and watch how decisions get made, rather than spend years on entry-level tasks that AI can increasingly absorb.
He framed this as a duty to younger workers. It’s also a sensible hedge. If the next generation never learns how experienced people weigh trade-offs, there will be fewer capable people to oversee agentic automation later.
What the Gulf Is Doing With Agentic Automation
The event also had a regional angle. Dines called the UAE government one of the most progressive he works with when it comes to adopting AI and software, and one of UiPath’s best partners.
Karl Crowther, who leads UiPath in the Middle East and Africa, opened the day. He has lived in the UAE for almost 20 years and described its people as resilient. He said the companies that kept core processes running through this year’s disruption were the ones that had automated first, and he likened automation to a shock absorber.
Crowther added that banks, government entities, retailers and oil and gas companies in the UAE are already discussing advanced agentic automation with UiPath. The source doesn’t say how far any of those talks have progressed, so they’re best read as a sign of interest rather than proof of large deployments.
Read Next
A few related pieces worth your time:
- Marvell Lumentum Acquisition Bets $115B on Optical AI Links
- Never Miss a Word Again: Best Wireless Headphones for TV in 2026
- Tech Startups Thrive with Strategic Finance Leases
What Leaders Should Take From It
It’s worth remembering that these remarks come from a vendor with a stake in the market. Even so, the practical points hold up without any sales pitch. Test what can be tested. Keep AI away from work that needs the same answer every time. Assign a named owner for anything an agent does on the company’s behalf. Those rules apply to any agentic automation program, whoever supplies the technology.
The people question may be the harder one. Retraining staff to take initiative and think in systems takes longer than buying software, and junior hiring plans need rethinking now rather than later. Companies that treat agentic automation as a staffing and governance project, not only a technology purchase, will be better placed to take on the responsibility Dines says can’t be handed off to the machine.
Source: The Next Web
