UiPath: AI agents force rethink of junior tech roles

The growing use of AI agents could fundamentally change how companies develop junior employees, with businesses potentially needing to move entry-level workers away from routine tasks and towards more strategic roles, UiPath CEO Daniel Dines has warned.

Speaking at an APAC media briefing coinciding with UiPath Fusion in Las Vegas, Dines said AI agents would increasingly handle operational work, but also argued that human initiative, accountability and customer relationships will remain important as enterprises adopt more autonomous systems.

One pertinent challenge for businesses going forward, according to Dines, is managing junior employees, who traditionally learn about their organisations through performing routine work and gradually gaining institutional knowledge.

"(Juniors) are probably the most AI literate among all the employees in a company," Dines said.

"They can instil a new way of working, however, at the same time, you don't want to hire a junior and give them manual tasks that agents can do."

Instead, he expects onboarding and career development to change, with those early in their career becoming involved in strategic decisions sooner and working more closely with senior employees.

The shift could also require changes to education systems, with businesses increasingly requiring system thinkers, rather than workers trained primarily to follow established processes.

Building a map for agentic deployment

Dines used UiPath's new 'map of work' concept to explain how the company believes enterprises should deploy agentic AI.

AI systems face a fundamental disadvantage compared with human employees because they do not naturally acquire the institutional knowledge that people develop when working inside an organisation.

Employees learn informal processes, exceptions, terminology and customer relationships over time, much of which is absorbed implicitly or through osmosis, not captured in formal documentation that AI can learn from.

Organisations, therefore, need a detailed representation of how work is actually performed before AI agents can reliably operate within those businesses.

UiPath is addressing this with the recently launched UiPath Cartographer, which is described as effectively performing the role of a business consultant.

The agent can analyse existing process maps, documentation and standard operating procedures, while also interviewing subject matter experts about how work is performed.

Dines said the agent could, for example, interview an accounts payable employee while observing how they process invoices, helping to capture knowledge that may otherwise remain in employees' minds.

The resulting map of work can then be incorporated into UiPath's platform, where it provides context for agents, automations and human decision-makers.

The approach is also intended to address one of the problems with traditional business process management: keeping an accurate representation of business processes as they change.

A process map can quickly become outdated as new exceptions emerge or regulations change.

UiPath's approach instead creates a feedback loop in which human decisions can be used to identify previously undocumented rules.

Dines gave the example of invoices above a certain value consistently being escalated for approval, even if that rule had never been formally documented.

By observing those exceptions, the system could identify a potential business rule, ask employees to confirm it and then incorporate the rule into the map of work.

Humans stay accountable

Dines also argued that companies should be cautious about allowing AI agents to operate autonomously, particularly when they can take actions with real-world consequences.

He said agents should initially operate with humans in the loop, with autonomy gradually earned based on demonstrated performance.

There is a clear and important distinction between an agent making a recommendation and being able to execute an action.

"If agents cannot call tools, there is zero risk," Dines said. 

"The moment an agent can call tools, make payments, do actions, this is where the real risk happens."

UiPath's approach involves assigning different levels of trust to the tools an agent can access.

An agent might initially be allowed to retrieve information or make recommendations, while actions such as payments or changes to business systems could require greater levels of approval.

It would eventually be given greater autonomy after demonstrating a high level of accuracy in a specific class of decisions.

Dines cited invoice processing as an example, suggesting that an agent consistently achieving 99 per cent accuracy on invoices below a particular value could eventually be allowed to handle those cases without human intervention.

However, he stressed the importance of new AI models being tested carefully because model changes can introduce regressions.

Business relationships still require human touch

While AI agents are expected to take on more operational work, companies should retain humans in areas where relationships, trust and accountability are important, with Dines pointing to customer service as an example.

A large medical equipment manufacturer recently discussed with Dines the potential to automate simple customer requests, such as checking an order status or locating an invoice.

However, that business also recognised that its customer service employees had developed relationships with customers over many years.

Those relationships have commercial value, even when the underlying customer service tasks are relatively straightforward.

Human employees would also remain necessary to take responsibility for decisions made by AI.

"An agent can make a mistake," Dines said. "Someone has to own the mistake."

This accountability becomes increasingly important as AI agents are given greater access to enterprise systems and business processes.

RPA to orchestration transition

UiPath is broadening its positioning beyond its traditional robotic process automation roots.

Dines said the company began developing its orchestration capabilities more than two years ago after recognising that enterprises were beginning to combine deterministic automation, AI agents and human decision-making within the same processes.

Rather than positioning itself purely as an orchestration provider, UiPath wants its orchestration, automation and agentic capabilities to operate in concert.

A single business process could involve a deterministic automated step, followed by a human decision, an AI agent performing another task and further automation.

The complexity of these processes makes an orchestration layer increasingly important, but orchestration should remain closely connected to the underlying automation capabilities.

"We don't want to play in orchestration alone," Dines said. "We are an orchestration and automation and agentic platform."