AI adoption in HR is moving into a more consequential phase. The question is no longer simply whether organisations should experiment with AI, but where it belongs in the flow of work, where human judgement must remain, and what it takes to move from scattered pilots to sustainable enterprise adoption. 

These questions came to the fore at a recent roundtable in Manila, held in partnership with HONO, which brought together senior HR, technology, and business leaders from banking, retail, healthcare, hospitality, manufacturing, and other industries to examine the changing role of HR in the AI era.

Setting the context, Amandeep Singh from HONO reflected on just how quickly the conversation has moved, from the emergence of generative AI to agents capable of executing transactions and increasingly autonomous workflows. The opportunity, he argued, is shifting from experimentation towards embedding AI more deeply into how work gets done. But this also raises harder questions about adoption, governance, privacy, security, and the economics of AI at scale. HONO also introduced its vision for a conversational, zero-UI HRMS via ERA, designed to enable employees to interact with HR systems in natural language rather than through traditional menus and navigation.

Across the discussion, three priorities emerged for HR leaders navigating the next stage of AI transformation.

Keep humans where context matters most

As organisations automate more HR activity, leaders were clear that the goal should not be to remove people indiscriminately from processes. Instead, HR needs to become far more deliberate about deciding where automation creates value and where human judgement remains essential.

Repetitive transactions, information retrieval, analytics and administrative work present obvious opportunities for AI. But employee relations, sensitive decisions, interpretation and moments requiring empathy remain different.

Participants pointed to examples ranging from labour relations and employee termination to banking and other trust-intensive environments. AI may be able to surface information, identify patterns or recommend an action, but the final decision can depend on context that is difficult to reduce to a rule: an employee's circumstances, organisational history, cultural considerations or the consequences of a decision.

That distinction becomes increasingly important as agentic systems move from answering questions to taking action. Human involvement cannot simply be added after something goes wrong; it must be designed into the workflow.

Leaders also stressed that humans remain critical before AI is deployed. They must define the business problem, provide clean and meaningful data, test outputs and determine whether an AI recommendation makes sense in the real-world context of the organisation. As one participant observed, the value of the human lies partly in making sense of information through experience and different scenarios before deciding whether to act on an AI recommendation.

The emerging model, therefore, is not human versus AI. It is a clearer division of labour: machines taking on repeatable execution and intelligence-heavy tasks, while people concentrate on judgement, empathy, accountability and business value.

Adoption starts with the work, not the tool

A second message from the room was equally strong: providing access to AI does not automatically create adoption.

The bigger challenge is helping employees understand how AI is changing their work.

Organisations represented at the roundtable are taking different routes. Some have created AI academies and formal learning programmes. Others are running prompt challenges, gamified learning initiatives, innovation days, hackathons or pilot groups. The intent is not simply to teach employees how a particular tool works, but to create enough curiosity and confidence for them to begin experimenting.

One organisation described revisiting job descriptions and asking employees to separate their activities into work that should remain purely human, work that could be augmented by AI and work that could potentially be automated. The exercise shifts the conversation from abstract AI literacy towards concrete workflow redesign.

This also surfaced an important distinction between AI literacy and AI fluency. Employees may know what generative or agentic AI is without understanding when to use it, which tools are appropriate, or how to apply them to an actual business problem. Fluency develops through repeated application.

That is why small experiments can matter. Prompt challenges, departmental prototypes, and employee-led use cases provide a relatively safe environment to test AI against the problems they understand best. Several leaders also highlighted the importance of involving HR, IT, data, privacy and business teams together rather than treating workforce AI readiness as HR's responsibility alone.

The lesson is particularly relevant for organisations facing employee anxiety around automation. Telling employees that AI will "enable" rather than replace them will only go so far. Adoption strengthens when people experience firsthand how technology can eliminate repetitive work and create space for higher-value contributions.

Move from ‘launching AI’ to making AI land

If experimentation characterised the first chapter of enterprise AI, the next challenge is scale.

The discussion repeatedly returned to a fundamental business question: what is worth scaling, at what cost, and with what controls?

AI introduces costs that organisations need to justify alongside expected productivity benefits. Leaders spoke about selective access to enterprise AI tools, the choice between building capabilities internally and buying them from technology partners, and the importance of cybersecurity, data confidentiality, and system integration.

The conversation suggested a pragmatic path: pilot, stabilise and then expand. Rather than beginning with an enterprise-wide deployment, organisations can prove a use case within a geography, function or employee group, understand adoption and cost, resolve integration issues, and scale only when the model has demonstrated value.

But successful scaling also requires organisations to examine the processes underneath the technology. Automating a poorly designed or outdated workflow simply makes the same problem move faster. Legacy policies may conflict, processes may have evolved without clear ownership, and the data required by AI may be fragmented across systems.

This is where governance becomes inseparable from AI strategy. The more autonomy organisations give to technology, the clearer policies, permissions, escalation paths, and accountability need to be. The roundtable's closing discussion illustrated this through something as practical as conflicting HR and finance policies: an intelligent system can flag the contradiction, but the organisation still has to decide which policy should govern the action.

Ultimately, the next phase of AI in HR will not be defined by how many AI tools an organisation deploys. It will be determined by whether those tools meaningfully improve how work gets done.

For HR, that creates a larger mandate: redesign processes rather than merely digitise them, build workforce confidence rather than simply provide access, establish guardrails without slowing innovation, and protect the human judgement that employees and businesses will continue to rely on.

The technology may be advancing quickly. The real transformation will depend on how deliberately organisations redesign work around it.