Kiren Kumar, deputy CEO at Singapore's Infocomm Media Development Authority (IMDA), argues that regulating AI too early is a mistake that stifles digital growth and hinders innovation before it fully develops. Instead of rigid legislation, IMDA treats regulatory trust as an economic asset, co-creating voluntary testing sandboxes with tech firms to guide behavior organically. Kumar emphasizes, "We don't believe regulating it right now is the answer." Singapore's approach leverages its global reputation for stability—built across decades in aerospace and semiconductors—as a foundation for positioning itself as a secure testing ground for emerging AI industries.
Regulatory Philosophy: Trust as Economic Asset
Singapore deliberately rejects both regulatory extremes. Rather than passing rigid laws, IMDA builds voluntary testing sandboxes to guide corporate behavior before rule-breaking becomes a crisis.
Kumar notes that Singapore's brand relies entirely on trust. The nation positions itself as a secure testing ground for emerging industries by working directly with companies to build governance frameworks. "Some countries regulate technology, others don't," Kumar says, pointing to IMDA's middle-ground approach.
Policy Translation: From Framework to Code
To make governance frameworks useful, policy must translate into actual code. IMDA launched testing tools like Moonshot, which allow developers to evaluate their models against governance frameworks before deployment. Results are then published to educate the global ecosystem.
Agentic AI: New Risks Require Rethinking Governance
This collaborative approach faces stress from the rise of agentic AI—autonomous software that executes multi-step plans without human approval. Kumar explains that because agentic AI can reason and take action with no human in the loop, it introduces new risks regarding safety and reliability that static laws cannot effectively address.
"With [agentic] systems, you're going to have multiple agents working together, and I think then we need to rethink how we frame the model governance framework," Kumar says, stressing that oversight must be built around multi-agent use cases.
Production Safety: Continuous Post-Launch Patching
Moving AI from pilot programs into live production is where errors become critical. Kumar expects and demands continuous post-launch patching. "The mental model is that there will be errors, there will be mistakes," he argues.
The key to survival is having a mechanism and business response ready to continually upgrade and tweak systems even after they reach the public. Connecting intelligent models to legacy databases is precisely where data leaks and security breaches happen. Kumar believes companies "need a sandbox" to ensure their data, architecture, and software connections are handled safely and reliably before putting systems into production.
He urges boards to treat software deployment like physical engineering: "Pilot to production is no different from how an engine manufacturer will test its engines before they put it on a plane."
Leadership and Talent: The Final Barriers
Hesitant executives and a global shortage of specialized talent remain barriers to AI adoption. "This is a leadership question," Kumar says, noting that an executive's drive to force organizational change matters more than government policy.
This gap is compounded by a lack of technical resources. Many mid-size and small companies understand their business domains but lack in-house teams to build and deploy custom AI solutions. As a result, "forward-deployed engineers are becoming a scarce commodity globally because they need to work hand in hand with the client, understand the workflow, and deploy the technology."
Singapore's Deployer Strategy
To overcome talent shortages, Singapore ignores the race to build frontier models from scratch. Instead, the nation imports global algorithms and deploys them into highly regulated industries.
Kumar argues, "We strongly believe that Singapore is positioned to be a deployer of these technologies at scale in a responsible, trusted way."
IMDA identified advanced manufacturing, finance, connectivity, and healthcare as prime targets. Because failure in these fields is costly, they demand a higher bar for trust, reliability, and human judgment.
Beyond Efficiency: Achieving Business Transformation
Surviving the AI shift requires more than minor cost-cutting. "A lot of these pilots are… [designed to] drive productivity by 10% to 20%… which is valuable. But how do we get to 10x?" Kumar asks.
Hitting that multiplier requires transforming the business workflow to create entirely new products and services.
Workforce Upskilling: From Theory to Daily Tasks
To achieve this transformation, technology must move out of the engineering department and into the hands of ordinary workers. Kumar argues that true economic value is unlocked only when everyday professionals—from lawyers to marketers to HR staff—are empowered to integrate AI into their daily routines.
To drive adoption, Singapore launched a national initiative to upskill 100,000 workers. Rather than offering abstract computer science classes, the program focuses on "online courses and certification for their particular workflows… It's on-the-job training; it's contextual, not theoretical."
This approach extends to final-year students, who are enrolled in the same programs as working professionals. The goal is to close the gap "and get them to be job-ready or AI-ready."
Broader Context
Kumar's caution against regulating AI too early reflects a philosophy that differs from the global regulatory direction. The EU's AI Act already sets binding, risk-based obligations for AI developers and deployers, while EU member states are required to establish AI regulatory sandboxes under the Act. This suggests sandboxes are useful as a complement to hard rules, not a substitute for legislation.
McKinsey's 2025 State of AI survey found that AI adoption is widespread, but most organizations are still struggling to move from pilots to scaled impact. Its 2025 workplace AI report found that just 1% of companies describe themselves as mature in AI deployment, suggesting that trust infrastructure matters, but leadership, operating models, data readiness, and workflow redesign remain bigger bottlenecks for many firms.
Kumar's focus on forward-deployed engineers also points to a constraint that policy cannot solve quickly. Business Insider reported in May 2026 that forward-deployed engineer job postings rose 729% over the previous year, reflecting surging demand for people who can translate AI into real enterprise workflows.
