In January 2026, Singapore's Infocomm Media Development Authority published the Model AI Governance Framework for Agentic AI — the first national governance framework specifically addressing autonomous AI agents. It was a significant moment in the AI governance conversation. Countries have been regulating AI in general terms for years, but this was the first framework to grapple directly with agents: AI systems that can take independent action in the world.
I read the framework with a strange sense of recognition.
Four Principles, Already Built
Singapore's framework articulates key governance principles for agentic AI: human oversight of agent actions, transparency in agent reasoning, accountability for agent outcomes, and secure handling of agent capabilities. These principles are sound. They reflect careful thinking about the risks and responsibilities inherent in deploying AI systems that can act autonomously.
They also describe, almost exactly, the governance mechanisms I had built into LadenX months before the framework was published.
Human Oversight
LadenX's human-in-the-loop approval gates require explicit human authorization before any dangerous operation executes. The AI proposes. The human decides.
Transparency
Full audit logging records every action, decision, and reasoning chain. Every step is traceable, reviewable, and accountable.
Accountability
The three-tier risk classification system and the clear chain of responsibility from AI recommendation to human approval establish unambiguous accountability.
Secure Capabilities
AES-256-GCM envelope encryption for credentials ensures the AI never handles raw secrets. Capabilities are secured at every layer.
I did not build these mechanisms because a governance framework told me to. I built them because I was an engineer with SSH access to production servers who understood, from hard experience, what happens when autonomous systems operate without adequate safeguards. The governance emerged from practice, not from policy.
Governance in Code
This is the gap I want to talk about. Engineers are solving AI governance problems in code every single day. Every developer who implements rate limiting is making a governance decision. Every team that builds an approval workflow for an AI action is creating oversight architecture. Every engineer who adds audit logging is building transparency infrastructure. Every practitioner who classifies operations by risk level is doing the work of governance.
But engineers — especially engineers from the Global Majority — are largely absent from the policy conversations that will determine how AI is governed worldwide.
The Representation Gap
The major AI governance initiatives reflect this absence. The NIST AI Agent Standards Initiative, launched in February 2026 to address autonomous AI systems specifically, was developed primarily through consultation with US-based institutions and corporations. The EU AI Act was shaped by European lawmakers, industry lobbies, and academic institutions. The African Union's Continental AI Strategy represents an important step, but the engineering practitioners building AI systems across Africa are underrepresented in these conversations.
The result is a gap between policy and practice. Governance frameworks describe principles. Engineers implement mechanisms. When the two are developed in isolation, both suffer. Frameworks become abstract. Implementations become ad hoc.
Practitioner Knowledge Matters
I've seen this from both sides. At Command Space, leading a team of seven, I make governance decisions in code daily — decisions about what our AI systems can and cannot do, how they handle sensitive data, how they interact with human operators. These decisions have real consequences for real users. They are informed by practical experience that no policy document can fully capture.
When I facilitated the AI workshop for the Stanford SEED Transformation Network in Accra — roughly 150 entrepreneurs engaging with AI tools for the first time — the questions were intensely practical. How do I know this AI won't misuse my data? What happens when the AI makes a mistake in my business? Who do I hold accountable? These are governance questions, asked by practitioners, that deserve practitioner-informed answers.
More Seats Needed
The AI governance table needs more seats. Specifically, it needs seats for the engineers and practitioners who are building the systems these frameworks aim to govern. And it needs seats for practitioners from the Global Majority — from Accra, from Lagos, from Nairobi, from Mumbai, from Jakarta — whose contexts, constraints, and contributions are currently underrepresented.
I am not a policymaker. I am an engineer who builds AI systems that operate in production environments. When Singapore published its framework and I saw my own engineering decisions reflected in national policy, it confirmed something I'd felt for a long time: the engineers building AI and the policymakers governing AI are working on the same problems from different directions. It is past time we started working on them together.