Walk into a modern tech hub or enterprise IT department in Southeast Asia today, and you’ll find AI coding agents becoming an increasingly common part of the developer workflow. This momentum is expected to continue, with Gartner predicting that 90% of enterprise software engineers will use AI code assistants by 2028.
In less than two years, coding agents have evolved from simple autocomplete assistants into autonomous programming tools. In a region facing tech talent shortages — where, according to General Assembly’s “The State of Tech Talent 2026” report, 95% of employers in Singapore alone face IT hiring challenges — the ability to use natural language to direct an AI agent to build software is compelling. However, realising the value of this capability requires more than just generating clean code. While an AI agent can quickly produce a clean script on a developer’s laptop, operationalising that code at scale is another matter entirely.
Beyond code generation
Industry conversations often focus on the initial creation phase. What we tend to overlook is that writing code only represents 10% of the journey. The remaining 90% is where the real work begins: the immense challenge of scaling applications in complex enterprise environments.
Consider the process of applying for a mortgage through a banking application. On the back end, a mortgage application involves dozens of moving parts: executing a credit pull, verifying income data, conducting automated document reviews, running underwriter checks on exceptions, and sending a final confirmation back to the applicant.
Every one of those steps relies on automation, and the entire experience only feels seamless to the customer if every handoff between those systems functions correctly. When a handoff fails, the customer experiences the friction directly, and the business may also be affected.
Supporting these processes requires an operational infrastructure that encompasses everything from background retry logic following an API failure to secure credential vaults and strict role-based access controls. None of these structural foundations is visible in a demo. Yet they distinguish a proof of concept running on a developer’s laptop from a workflow operating reliably across dozens of business units.
The quality assurance bottleneck
Increasingly, the software development bottleneck is shifting from code generation to validation. Traditional testing approaches may struggle to keep pace with the influx of AI-generated code, which can exceed the capacity of manual human review. Expecting teams to manually inspect and verify these volumes of machine-generated code may no longer be a viable strategy for meeting delivery timelines.
Testing can no longer function as a downstream checkpoint performed after development is complete. It must become an integrated component of the software lifecycle, operating continuously alongside development.
Yet testing is only one piece of the equation. As enterprises deploy growing numbers of AI agents across development, operations, and business processes, they need a consistent way to govern, monitor, and coordinate increasingly complex systems.
The orchestration imperative
This is where orchestration comes in. Rather than focusing on a single coding model or development tool, organisations may require an underlying control layer that provides visibility, governance, and execution management across the software lifecycle. Whether teams are using Claude Code, OpenAI Codex, or future generations of coding agents, an orchestration layer can connect these systems and help them operate within defined business and security boundaries.
Governance is particularly critical in highly regulated industries such as financial services, where institutions face increasing scrutiny over operational resilience, auditability, and AI usage. Security controls, policy enforcement, credential management, audit trails, and role-based access controls cannot be treated as optional add-ons. They must be embedded into the development process itself, ensuring that software generated by AI agents follows the same governed pathway to production as software created by human developers.
At the same time, advances in autonomous and self-healing testing are changing how quality assurance functions operate. Rather than manually validating every line of code, testing teams increasingly work alongside AI agents that continuously execute tests, identify issues, and adapt to application changes throughout the software lifecycle.
Human oversight, however, remains important. As AI agents take on greater responsibility for code generation, testing, and maintenance, technology leaders are increasingly moving towards a human-on-the-loop model. Autonomous agents perform the operational work, while human experts retain accountability for governance, risk management, and final release decisions.
As coding agents become increasingly capable, the ability to generate code quickly may become standard. Organisations will also need to govern, test, and operationalise AI-generated software at scale.
Effective deployment will depend on whether organisations can manage AI-generated software securely and reliably.
