More than half of developers across Southeast Asia and India are already running AI agents in production workflows. Yet fewer than four in ten say their codebase is ready for fully autonomous operation.
That gap, documented in the Agoda AI Developer Report 2026, sits at the center of a structural shift now reshaping software development across seven of Asia’s most active technology markets.
The report, covered by The Tribune, was commissioned by Agoda and conducted in partnership with Macramé Consulting, drawing on developer input across Indonesia, Malaysia, Thailand, the Philippines, Singapore, Vietnam, and India.
Where Autonomous Agents Are Landing Fastest, and Where Governance Falls Short?
Paulius Pazdrazdys, a full-stack programmer turned AI and software observer, sees the adoption-readiness split as a pattern that extends well beyond enterprise development shops.
With 53% of surveyed developers reporting that AI agents are already embedded in selected or broad production workflows, and only 38% describing their codebase as mostly or fully ready for fully autonomous operation, Pazdrazdys argues the governance shortfall is appearing wherever data-intensive software is absorbing these tools quickly.
He points to consumer-facing, prediction-heavy software as an illustration of how the same dynamic plays out outside the enterprise context.
A platform like www.stakehunters.com leans on the same agentic models to crunch match data and generate forecasts, and it inherits the identical governance, data-security, and oversight questions the report flags for enterprise codebases. The category is not insulated from the gap simply because it serves consumers rather than internal users.
“What began as a way to speed up individual tasks is becoming a broader change in how software is built. Across industries, developers are seeing stronger productivity gains and are beginning to use agents across more of the development lifecycle. But greater capability brings new responsibilities.”
Pazdrazdys’s observation reinforces a core tension the report surfaces, that deployment velocity and oversight maturity are not moving in step.
Productivity Gains Are Real, But Human Review Remains the Norm
The productivity numbers in the report are striking. More than half of respondents (55%) now save at least seven hours each week using AI tools, up from just 18% in 2025. That is a substantial shift in a single year, one that signals AI assistance has crossed from occasional utility to routine dependency across much of the developer workforce surveyed.
Code quality is moving in the same direction. Some 62% of respondents say AI-generated code is usable often or almost always without major changes, a figure that suggests the output of these tools has matured considerably. Agents are now being applied across the full development lifecycle, from writing and reviewing code to testing and deployment tasks.
Yet the headline productivity story carries an important qualification. Despite the breadth of AI use, 86% of respondents still review or validate AI-generated outputs always or most of the time. Efficiency gains and human oversight are, for now, coexisting rather than trading off against each other.
The survey draws on developer communities across Indonesia, Malaysia, Thailand, the Philippines, Singapore, Vietnam, and India, making its findings among the more geographically comprehensive in the region.
Cost Leads a Cluster of Barriers Slowing Wider Agent Deployment
If productivity is rising, the constraints on going further are also clearly documented. Cost is the single largest barrier to wider agent adoption, cited by 28% of respondents. Integration complexity follows at 24%, with lack of governance trailing at 19%. None of these barriers is dominant enough to explain the gap alone.
Together, they sketch an environment where organizations are absorbing tools faster than they are building the financial, technical, and structural conditions to support them at scale.
Usage limits compound the picture. Four in five respondents say they are currently operating under some form of AI usage restriction, whether token quotas, usage caps, or budget ceilings.
That figure means constrained access is close to universal among the developers surveyed, not a minority experience. Organizations are deploying agents broadly while simultaneously rationing the compute those agents can consume.
One safeguard appears to be holding. Some 79% of respondents require human approval before AI-generated work reaches production deployment. That number signals that organizations are not yet willing to hand final authority to autonomous systems, even as they integrate those systems deeper into the development process.
Junior Developers Feel the Career Pressure; Leadership Frames It as Accountability
The career dimension of AI adoption is registering differently across seniority levels. Among junior developers, 49% report feeling less secure about their future career prospects in an AI-driven environment. Among CTO and VP Engineering respondents, that figure drops to 17%. The gap is wide enough to suggest that senior leaders and junior contributors are not looking at the same transition.
The skills the report identifies as increasingly critical offer some context for that unease. System architecture and design, AI literacy, and agent orchestration are emerging as the competencies most likely to define value in a developer’s role going forward.
These are skills that often come with experience, which may partly explain why senior figures feel more settled about what comes next.
Idan Zalzberg, Chief Technology Officer at Agoda, offered the clearest articulation of where the real difficulty lies. The study was conducted in partnership with Macramé Consulting, and Zalzberg’s framing at its release pointed past the productivity argument to something harder to quantify.
“The real cost is not simply the model call. It is the work required to make AI trustworthy in production. The organizations that will excel in this next phase will be the ones that manage cost carefully, build the right safeguards, and keep human judgment where it matters most.”
That framing leaves the accountability question open rather than resolved. Making AI trustworthy in production is not a single milestone. It is an ongoing discipline, and by the report’s own figures, most organizations across Southeast Asia and India are still working out what that discipline requires.
