Workplace Learning in the Era of Agentic AI: From Courses to Continuous Performance

Agentic AI—systems that can plan, take actions, and learn from feedback—changes the centre of gravity for workplace learning in Singapore. Instead of funnelling employees into periodic courses, learning must be integrated into the flow of work and tied to measurable outcomes, such as productivity, customer satisfaction, and cycle time. That shift is not just fashionable jargon; it's a strategic response to a job market where skills needs are evolving faster than curricula (WEF, 2025). Singapore's Skills Demand for the Future Economy report already points to rising demand across digital, care, and green economies, coupled with domain-specific skill sets and transversal capabilities like problem-solving and collaboration (SkillsFuture Singapore, 2024/25). If learning continues to sit outside the job, organisations will train, but workers won't transform.

Agentic AI makes "learning in the flow of work" a reality by embedding retrieval-augmented guidance, micro-coaching, and automated feedback into daily tasks. A sales associate can receive real-time recommendations for cross-sell language; a port planner can prompt an AI agent to surface schedule anomalies and get a step-by-step explanation of the chosen plan; a public sector officer can ask a safeguarded assistant for regulatory precedents and receive citations to primary sources. The point is not the chatbot—it's the closed loop: initiate a task, receive targeted help, perform the task, receive instant feedback, and update the playbook. This echoes the "learning in the flow" paradigm, long advocated by learning strategists, but now with the addition of automation and context awareness (Bersin, 2018; Wengel et al., 2019).

What holds companies back? First, governance. Agentic systems must be explainable, privacy-preserving, and aligned with corporate risk appetite. Singapore's PDPC guidance clarifies how personal data may be used to train or deploy AI and what consent or safeguards are required, offering a practical path for L&D teams who want to analyse performance data without tripping compliance (PDPC, 2024). In parallel, the Model AI Governance Framework for Generative AI outlines controls across data, the model lifecycle, safety, and accountability—useful scaffolding for "AI-inside-L&D" programmes (AI Verify Foundation/IMDA, 2024). Second, measurement. Many L&D dashboards track completions; few track work improvements. Singapore's recent productivity uptick is encouraging (MTI, 2025; SingStat, 2025), but organisations must attribute gains to specific learning interventions and agentic workflows to keep investment disciplined.

A practical playbook for Singapore enterprises:

1. Anchor on high-value use cases. Start where errors are costly or time-to-competency is long: financial advisory suitability checks, logistics slotting, healthcare patient flow, or precision maintenance. Tie each use case to a KPI that owners already care about (e.g., first-time-right rate), not "learning hours".

2. Instrument the job. Capture task context and outcomes (with privacy safeguards) so agentic assistants can see what "good" looks like. Utilise the retrieval of your SOPs and policies; include a feedback button in the UI to improve guidance continuously.

3. Codify governance by design. Align with the GenAI governance framework's controls (data lineage, red-teaming, incident response) and PDPC's expectations on consent and transparency. Provide human-in-the-loop escalation for sensitive decisions (IMDA/AI Verify Foundation, 2024; PDPC, 2024).

4. Upskill managers as "workflow learning" coaches. They don't need to be prompt engineers; they need to set task briefs, review AI outputs, and coach judgment. Research shows development opportunities are now central to employee experience—managers are the fulcrum (Wengel et al., 2019).

5. Close the ROI loop. For each use case, define a baseline (e.g., average handling time) and run A/B pilots comparing agentic workflows with the status quo, where improvement is proven. Scale and refresh SOPs so that the new way becomes the standard.

Agentic AI won't eliminate structured courses; it will right-size them. Regulatory and safety training will remain classroom- or e-learning-anchored. But the centre will be the job itself, supercharged by context-aware guidance. In Singapore's skills-hungry economy, the winners will be firms that operationalise learning as a living system inside work, governed and measured with the same rigour as any production asset.

References

AI Verify Foundation & IMDA (2024)

Model AI Governance Framework for Generative AI. Available at: aiverifyfoundation.sg (accessed 5 Oct 2025).

Bersin, J. (2018) 'A new paradigm for corporate training: Learning in the flow of work', Josh Bersin. Available at: joshbersin.com (accessed 5 Oct 2025).

Ministry of Trade and Industry (2025) Economic Survey of Singapore 2024 (released Feb 2025). Available at: mti.gov.sg (accessed 5 Oct 2025).

PDPC (2024) Advisory Guidelines on the Use of Personal Data in AI Systems. Available at: pdpc.gov.sg (accessed 5 Oct 2025).

SkillsFuture Singapore (2024/25) Skills Demand for the Future Economy. Available at: jobsandskills.skillsfuture.gov.sg (accessed 5 Oct 2025).

Wengel, J. et al. (2019) 'Making learning a part of everyday work', Harvard Business Review, 19 Feb. Available at: hbr.org (accessed 5 Oct 2025).

World Economic Forum (2025) Future of Jobs Report 2025. Available at: weforum.org (accessed 5 Oct 2025).