AI agents are moving from experiments into production software workflows across Southeast Asia and India, but most organisations are keeping people in control of consequential changes, according to Agoda’s AI Developer Report 2026.
The company says its survey covers more than 800 developers and engineering leaders in the region. Of those respondents, 53% reported that AI agents were already used in selected production workflows or more broadly across their organisation. The figure rose to 58% among respondents in Singapore.
Productivity gains are rising, but they are self-reported
Agoda says 55% of respondents now save at least seven hours a week by using AI, compared with 18% in its 2025 study. The sharp increase suggests coding tools are taking on a wider range of work than code completion alone, including multi-step tasks across development workflows.
These figures are based on respondents’ own assessments rather than independent time-and-motion measurements. They therefore indicate how surveyed developers perceive the effect of AI on their work, not a guaranteed productivity gain for every team or project.
The report’s examples extend beyond conventional coding assistants. Agoda says its CodeMaster platform gives coding agents a governed environment in which to work, with a target for agents to write 10% of the company’s merge requests this year. Other case studies on the report site cover uses such as payment-specification analysis, customer support, research and automated reporting.
Full autonomy remains the exception
Production adoption does not mean organisations are ready to let agents operate without supervision. Only 38% of respondents considered their codebase ready for full AI autonomy, while 79% said a human must approve an AI-led production deployment.
That distinction matters because generating a plausible code change is only one part of software delivery. An agent also needs suitable repository context, reliable tests, controlled access to tools and credentials, auditability, and a clear process for reviewing failures. Legacy systems and poorly documented codebases can make those requirements harder to satisfy.
The survey points to a model of bounded autonomy: agents can plan and perform more work within a controlled environment, but people remain accountable for approval and production outcomes. That approach may be particularly important where software affects payments, personal data, critical services or regulated operations.
Cost overtakes integration and governance as the leading barrier
Cost was the most frequently cited obstacle to wider adoption, selected by 28% of respondents. Agoda places it ahead of integration complexity and governance in this year’s findings.
The expense of an agentic system can extend beyond model tokens. Organisations may also need to pay for evaluation infrastructure, observability, sandboxed execution, security controls and the engineering work required to connect an agent safely to existing systems. Longer-running agents can multiply those costs as they search repositories, call tools and retry tasks.
For technology leaders, the findings suggest that adoption decisions should be based on the value of completed, reviewed work rather than the volume of generated code. A cheaper model that needs extensive correction may not deliver a lower total cost, while a capable agent may still be uneconomical for routine tasks that simpler automation can handle.
What the survey does—and does not—show
Agoda’s report provides a regional snapshot of how surveyed developers and engineering leaders describe their use of AI. The published summary does not establish that the sample represents every developer or organisation across Southeast Asia and India, and the results should not be read as independent proof of a particular tool’s performance.
Even with those limits, the gap between production use and readiness for full autonomy is instructive. AI agents are becoming part of everyday engineering work, but codebase quality, operational safeguards, cost controls and human accountability remain central to deploying them responsibly.
The full Agoda AI Developer Report 2026 includes country-level findings and additional case studies.
Source: Agoda AI Developer Report 2026 and accompanying press information distributed by WE Communications on 25 September 2026.
