Building Prompt Engineering & Agentic AI Readiness at Enterprise Singapore (AAD)

The Challenge: Building Practical AI Skills Within Government Restrictions

AAD officers were already using AI tools such as ChatGPT and individual Claude accounts in their daily work, but most were not yet taking advantage of the features and techniques that make the biggest difference. The team recognised that prompt engineering alone would not change how officers work day to day, and that the real skill gap was agentic AI — where officers could see what genuine task automation looks like.

The setting brought its own realities. As a government division, AAD operates within clear restrictions on tool choice and data handling. File uploads are restricted, non-whitelisted tools cannot be used on government laptops, and sensitive accreditation, compliance and applicant data cannot be brought into any external platform. During planning, one preferred tool could not be accessed on office devices at all, so the workshop had to be built around approved, browser-based tools instead.

AAD therefore needed a session calibrated to its approved tools, grounded in real accreditation and assurance work, and designed so officers could learn safely without ever needing to expose sensitive data.

Why Heicoders Academy: A Partner Who Designs Around Real Constraints

As Singapore's leading AI-First Skills Accelerator, Heicoders Academy has delivered Generative AI training to hundreds of organisations across the public sector, financial institutions and large enterprises. That depth of experience meant the curriculum could be adapted to a regulated government environment rather than delivered off the shelf.

When it became clear that the initially preferred platform could not be used on AAD's office devices, Heicoders pivoted the delivery to ChatGPT, aligning with the ChatGPT Enterprise and ChatGPT Agent access being rolled out to officers. All exercises were designed around generic, non-sensitive sample scenarios drawn from accreditation reviews, compliance assessments and assurance reporting, so the workshop could proceed safely regardless of final device or tool decisions.

The Solution: A Customised Full-Day Workshop

Heicoders Academy delivered a full-day, six-hour workshop on-site for 30+ AAD officers on 11 August 2026. The curriculum, partially adapted from the flagship GA100: Generative AI for Automation and Productivity programme, progressed deliberately from prompt engineering fundamentals to hands-on agent building across six focused segments:

• Introduction to Generative AI: How large language models work and power everyday chatbots, reasoning and thinking models, memory, custom instructions, and AI-powered media creation including image and voice generation.
• Prompt Engineering Fundamentals: Core prompting techniques applied to accreditation and assurance scenarios such as drafting assessment summaries and compliance review notes, plus practical strategies to minimise AI hallucinations through data grounding, negative constraints and mandatory citations.
• Practical AI Applications for Accreditation and Assurance Work: Using AI deep research to review regulatory standards, drafting assessment reports and stakeholder correspondence, data analytics and visualisation, and building custom GPTs for repeatable tasks.
• Introduction to AI Agents: The difference between using AI as a tool through prompting and AI as a system through agents, and what makes an AI agent genuinely agentic.
• Utilising AI Agents to Automate Workstreams: A hands-on lab using ChatGPT Agent, where officers executed real workflows including research and staff paper writing, compliance and accreditation data analysis, scheduled automated tasks, spreadsheet automation and slide generation.
• AI Governance: Principles of AI governance as a decision-rights framework, key watch areas such as hallucination, bias and prompt injection, and recommended policies for safe organisational AI use.

Throughout the day, every demonstration and exercise used sample, non-sensitive scenarios, so officers could build real skills without exposing accreditation or compliance data. Because the underlying prompt engineering and agentic AI concepts transfer across ChatGPT, Claude and other tools, the skills officers gained will continue to apply as their tool access expands.

Results: A Practical Foundation for Agentic AI

By calibrating the content to approved tools and grounding it in familiar accreditation and assurance work, the workshop gave officers a concrete, hands-on understanding of what agentic AI can do, rather than abstract theory. Officers moved beyond basic chatbot usage and, for many, built and ran their own AI agents for the first time.

The session was designed as an exposure and readiness step. As agentic tools become more integrated into the government ecosystem, officers who now understand these concepts are better placed to adopt new capabilities as access expands.

What's Next: Deepening AI Capability Across Enterprise Singapore

With a practical foundation now established within AAD, both teams are well positioned to explore more in-depth, hands-on training that builds on this workshop.

This engagement also sits within a broader, growing relationship between Heicoders Academy and Enterprise Singapore, and both organisations look forward to exploring how AI capability can be developed further across more teams in the months ahead.