Towards an HR function augmented and orchestrated by agents

The time has come to rethink the HR function from top to bottom. The novelty lies not only in generative AI, but also in the rise of "agents" capable of understanding context, reasoning, planning sequences of actions, and connecting to business tools to execute tasks end-to-end (our AI Agents Guide (This mechanism is detailed). In concrete terms, we are moving from incremental functionalities integrated into HR software to a transformational approach, where investment in agentic AI unlocks greater value potential by automating, personalizing, and empowering decision-making at scale. It is this shift—from integrated AI to agentic AI—that redefines the HR value proposition for the benefit of both the company and its employees (see the value/investment curve, p. 2).

This shift is resulting in a gradual shift in work, along a clear continuum: "AI-assisted" (moderate support), "AI-augmented" (human-machine co-management with handovers), and "AI-powered" (execution mostly carried out by agents, with human supervision). Examples abound: answers to employee questions, HR transactions, analyses, and reports are natural candidates for transfer to agents, while strategic consulting, talent strategy definition, or human support activities remain predominantly human, but amplified by insights produced by AI (p. 3). The target is not blind automation, but an intelligent reallocation of human effort toward value.

What makes an agent more than a sophisticated "chatbot"? A reasoning core capable of understanding, planning, executing, correcting itself through iterative reflection, and collaborating with other agents. An agent knows how to extract intent, develop a plan, call external tools, verify its results, and learn from feedback. In multi-agent systems, it also knows how to coordinate, delegate, and exchange information to accomplish complex tasks that exceed a single role (agentic capability diagram, p. 4). The result: assisted or autonomous operations that infuse intelligence into processes, IT, and ultimately, the employee experience.

The promise is concrete: accelerate the production of actionable insights, integrate internal and external data in real time, collaborate seamlessly with applications and humans, while maintaining a traceable audit of decisions and a corporate control framework (p. 5). This creates a virtuous circle of continuous learning and innovation, where “intelligence” is not a one-off deliverable but a living mechanism that improves “people” journeys, policies, and products.

To seize the right opportunities, you need to start with the desired business outcomes, then work your way up the chain of actions, the necessary skills (analyze, compare, plan, evaluate, generate, etc.) and the knowledge to be mobilized (terminology, processes, meaning of data). It is the pyramid "Outcomes → Actions → Abilities → Knowledge" that helps identify use cases where decision-making, knowledge and reasoning are at the heart of the value (p.6). In practice: we target situations where a series of tool-based actions leads to a measurable result - for example, reducing the integration time of a new employee or anticipating a risk of departure.

This method sheds light on the transformation of roles. On the HR Business Partner side, freed from administrative tasks by assisted/augmented AI, we shift from "firefighter" reaction to data-driven proactivity. HRBPs can orchestrate the integration of AI into businesses, detect weak signals about skills and organizations, and coach managers and executives on a larger scale thanks to augmented capacity and a coherent knowledge base (p. 8). Sustainable human capabilities—data acuity, critical thinking, emotional intelligence, negotiation—become the differentiator, boosted by AI fluency.

Centers of Expertise, for their part, are moving from a fragmented, manually updated model to a “product” model driven by dynamic insights. Content and policies are synchronized with internal and market trends; responses to employees become personalized and timely; the continuous improvement loop is based on large-scale listening and agent learning (human-machine collaboration diagram, p. 9). The challenge is no longer simply to “write a policy,” but to maintain a living system that adjusts, measures, and improves the experience.

In HR operations, the shift is even more dramatic: most administrative work is being performed by AI, while teams focus on monitoring, trend analysis, content enrichment, quality, and resolving cases requiring a human touch (p. 10). Recurring reports, prioritization, and dissemination are generated and distributed by agents; operations become a cockpit for experience optimization, not a ticket factory.

For employees and managers, the benefit is broken down into eight families of interactions: access to the right support, personalized answers to questions, access to personal data, simple transactions without logging into back offices, concierge support on complex journeys (onboarding, mobility), proactive nudges, explained analytics for better decision-making, and co-creation of content with an assistant (table of use cases and adoption levels, p. 11). The fastest adoption concerns support and information search; complex transactions and personalized analytics follow as the architecture matures.

To achieve this, we need to “rewire” human-machine collaboration. We move from a linear model where humans assign tasks to machines and wait for them to “finish” to a continuous loop where AI proposes, tests hypotheses, learns, and makes autonomous decisions within a governed perimeter, while humans retain control of ethical judgment, creativity, and strategic choices (model comparison, p. 12). This loop shortens the time between signal, action, and result—and raises the level of human intervention.

Concrete scenarios illustrate this approach. Upon returning from leave, an agent detects the upcoming event two weeks before the deadline; they plan a workflow adapted to the employee's context (type of leave, country, role), notify the manager on their preferred channel, provide a personalized checklist, orchestrate updates in the systems, and close the process before collecting feedback to improve the system (pipeline "detect → plan → nudge → insight → action → learning", p. 13). We no longer "execute" a process: we support a human from start to finish, with reproducibility and finesse.

Another key case: turnover prevention. An agent identifies a high-performer at risk by cross-referencing seniority, job market, and compensation benchmarks; they suggest options adapted to the manager's budget, execute the chosen recognition, and learn from the feedback to refine their recommendations (p. 14). The key point is not the scoring itself, but the ability to transform the insight into contextualized action, within the window of opportunity.

Behind these examples lies a useful taxonomy of agent types—reactive, proactive, adaptive, collaborative, autonomous, intelligent, and “social”—and a mapping of their opportunities across HR domains (p. 17). In the short term, there is a high potential for reactive/proactive/collaborative agents in talent acquisition, operations, and experience management; in the longer term, autonomous and “social” agents will become more prominent, particularly in training and cross-functional orchestration.

The "heatmap" of HR opportunities shows the breadth of the field: from passive sourcing to pre-selection, from knowledge management to compensation, from employee relations to surveys, from time & attendance to payroll, from succession plans to strategic workforce planning (p. 18). It is not a question of "automating" everything at once, but of structuring a portfolio: "AI-powered" cases where the agent leads; "AI-augmented" cases where the agent co-manages; "AI-assisted" cases where the agent accelerates and secures without deciding.

Technically, three architectural patterns coexist and will hybridize: the traditional API; the “agent-to-agent” where an agent talks directly with another agentic service; and the multi-agent system (MAS), which will become dominant on the user experience side within 12–18 months (p. 19). The target stack includes an experience layer (web, chat, mobile), an AI/orchestration layer with reasoning engine and responsible governance, a services/applications layer (HCM, CRM, LMS, case management, etc.) and a unified data/analytics layer.

Onboarding illustrates this orchestration: a “New Hire Agent” captures intentions (“sign up for benefits”, “complete my training”), triggers the right flows via the orchestrator, collaborates with a “Benefits Enrollment Agent” and a “Learning Plan Agent” who schedule, advise and attest, based on data from Core HR and knowledge repositories (end-to-end diagram, p. 20). The value is not only the reduction in time-to-productivity, but the elimination of experience gaps.

At the enterprise level, implementing a MAS ecosystem requires a trusted architecture: humans “in the loop” at control points, agents with memory and tools, workflows that orchestrate humans and machines, API/event-driven integrations, and a “models, tools & security” layer that covers mini- and large models, automation, private fine-tuning, governance, compliance, and cybersecurity (p. 21). All of this is based on a robust data fabric and a hybrid AI platform, not to mention the “business process layer” where business transformation lives.

How to act now? First, define three "North Star results" linked to experience and efficiency metrics (for example: employee NPS on life moments, T1 resolution time, self-service adoption rate). Then, run an "Outcomes → Actions → Abilities → Knowledge" diagnosis on two or three priority paths (onboarding, internal mobility, payroll support), and select a mix of AI-powered/augmented/assisted cases. Finally, define the cross-functional "enabler": an HR knowledge agent connected to the policy/procedure repository and usage signals, which self-updates.

Success also depends on change management. HR capacity must be created by shifting 60–80% of repetitive administrative tasks to agents to reinvest human time in coaching, designing data-driven policies, and supporting managers. The entire HR department must be trained in AI fluency (writing useful prompts, reading explanations, auditing recommendations), implementing usage guidelines, and establishing review rituals where the usefulness, safety, and fairness of agents are assessed.

On the governance side, three principles are non-negotiable: traceability (logging of decision paths and executions), controllability (guardrails on action scopes, autonomy thresholds, conditional human approvals) and accountability (monitoring of quality, bias, latency and cost). Agent "supervision" is becoming a full-fledged role in HR operations: we track metrics of accuracy, satisfaction, efficiency and adoption, and we feed the continuous improvement loop with interaction data and user feedback.

On the technological side, a pragmatic path is to start with a conversational overlay connected to key systems via an orchestrator, then gradually introduce specialized agents (leave, benefits, training, reporting) and their collaborations, while consolidating the data layer (HR glossary, policies, case histories, collaboration signals). The inflection point occurs when we can deploy “detect → plan → nudge → insight → action → learn” patterns across multiple life moments — the benefits become cumulative and visible.

Finally, it is essential to keep human ambition at the center. Agentics only makes sense if it expands access to quality support, makes the experience more personalized, and allows HR professionals to fully play their strategic role: anticipating, connecting, and illuminating. The alliance of machine reasoning and human judgment is not a compromise; it is the condition for an HR function that is fairer, faster, and closer to the realities of teams. The next generation of HRIS is not software: it is a living system of agents and humans who solve, learn, and improve together.

To learn more: discover how the AI agents in HR are already settling in, Which jobs will become automated or reorganized? and what our survey on the impact of AI on employment.