AI has not changed what good work is at Agoda. It has changed how people learn it
- Josephine Tan
Most organisations can now account for what AI has given them. Hours returned, drafts produced, tickets closed, cycles shortened. Far fewer can account for what it has quietly taken away.
For Paolo Inga, Chief Organisation and People Officer at Agoda, the second column is where the risk sits. “One of the biggest risks is that organisations and teams start treating AI output as an answer rather than a starting point,” he tells HRM Asia.
What erodes is not output but the capability underneath it. “If people accept it without questioning, its assumptions, checking the facts or considering the context, they risk weakening the critical thinking and subject-matter knowledge that help them spot mistakes and make sound decisions,” he says. The damage is slow, and largely invisible while it is happening: “That may not show up in efficiency metrics at first, but over time it can affect the quality of decisions and people’s ability to handle unfamiliar problems.”
Research points in the same direction. A study by Microsoft and Carnegie Mellon University, drawing on 319 knowledge employees, found that higher confidence in generative AI (GenAI) was associated with less critical thinking, while higher confidence in one’s own expertise was associated with more. Those least certain of their own judgement were the likeliest to take what the machine produced at face value.
Agoda’s response has been to draw a line through the work itself. “’Move fast and be efficient’ means using technology to reduce the ‘boring’ work like aligning boxes on a slide – not outsourcing the thinking,” says Inga. Structured learning runs alongside practical experimentation, so that employees learn not only how to use the tools but how to interrogate what comes out of them.
Protecting the learning, not the task
The problem lands hardest on early-career employees, whose training ground has long been precisely the work AI now does fastest. Research from Stanford University published last year found a 13% relative decline in employment among US employees aged 22 to 25 in the occupations most exposed to AI, even as employment held up for their older colleagues in the same roles.
Inga starts by separating the task from what it taught. “The goal should be to protect the learning that used to sit inside that work, not the task itself,” he says.
He uses the slide deck as his example and is unsentimental about it. “That was never necessarily the best way to learn; it was simply one of the available ways junior employees learned structure, context and judgement,” he says. If AI prepares the deck, the work moves up a level, to “thinking critically about the output, putting themselves in the position of the person receiving the presentation, and improving the output from that point of view”.
The expectations shift; the foundations do not. “The technical skills we expect from juniors are shifting with AI, but the fundamentals have not changed,” Inga says. “Curiosity and critical thinking matter as much as they ever did, and this is something we emphasise a lot in both the hiring process and people development.” One line carries the whole argument: “You can only judge AI’s work if you understand the strategy and context yourself.”
In practice, much of that falls to managers, who, he says, should be asking what AI produced, what was changed, what was accepted, and why. Agoda ran its 2026 summer internship programme over 10 weeks with more than 85 interns, who learned to use the organisation’s AI tools and presented what they built to department leaders; a separate Tech IC Foundations programme sets out for early-career engineers what their role involves and how they progress. “AI can automate some tasks, but it does not replace mentorship,” says Inga.
His deeper claim is that the standard itself has not moved. “AI has not changed how we tell good work from bad. Humans can also produce weak work, and learning to spot it has always been part of building judgement,” he says. “What is different is the speed. AI can hand you a polished answer in seconds, but someone still has to ask whether it is right, what is missing and whether it fits the situation.”
Agoda’s engineering teams, he notes, work with tools that will not give the same answer twice, which makes both the instruction and the review consequential. “I tell my team it is fine to use AI output, but they need to be able to show they have thought critically about the output, understand it, and stand by it,” he says. “The ones who stand out are curious about how they got to the answer, and not only about the answer itself.”
For that to hold, people have to be able to get things wrong in front of someone. Leaders, HR and managers, Inga argues, need to create the space for employees to explain their thinking, take feedback and make mistakes, including when using AI.
The organisation’s own figures suggest the habit is forming. In Agoda’s 2026 AI Developer Report, 86% of developers surveyed across South-East Asia and India said they review or validate AI-generated output always or most of the time, and 79% said human approval was required before production deployment.
Designing for the tools as well as the people
Asked whether AI changes how work itself should be structured, Inga does not reach for team shapes or reporting lines. He goes to what a team is now made of. “I think AI adds another dimension to organisational design,” he says. “Alongside the people in a team, we now need to think about the AI tools and agents supporting them, and how that combination helps each person achieve more.”
The practical expression of that is budge6ing. Inga points to NVIDIA CEO Jensen Huang’s suggestion that engineers be given an annual AI token budget alongside their pay. Agoda gives teams visibility into their own usage and costs, provides shared tools, and sets spending caps against the work being done – room to experiment, with the price of the experiment in view.
He expects the arrangement to become unremarkable. “Over time, providing that support could become as standard as giving employees access to Microsoft 365 or Google Workspace,” he says. For HR and business leaders, that converts into a set of design questions: which tools a team needs, how people work alongside agents, and where responsibility sits once they do.
On the last of those, Agoda’s position tightens. Experimentation is encouraged across functions, under an inside-out approach that builds capability internally first, but usage guidelines firm up as soon as an experiment moves towards production. The person using the tool is responsible for setting the instructions and parameters, checking what comes back, and deciding what happens next.
Which returns the argument to where it began. “Managers still set direction and develop people, and employees remain accountable for the work they produce,” says Inga. The tasks are moving; what a capable employee is expected to be able to do with them is not. “The task may change,” he says, “but the development of judgement still has to be intentionally designed from day one.”
