Agoda engineering blog

Agoda’s 2026 developer report examines how agentic AI is changing software development across Southeast Asia and India. It highlights adoption, readiness, hidden costs, human accountability, and practical controls organizations are using to deploy coding and operational agents safely.

Agoda describes how it evolved a Slack-based engineering support workflow from structured forms into conversational AI agents that gather context, route requests, investigate incidents, and escalate to humans. The post covers domain-specific agent platforms, tool integrations, tracing, evaluation datasets, and early results including 72.3% thread coverage and reduced supporter effort.

Agoda describes its migration of a terabyte-scale hotel price cache from a 72-shard SQL Server deployment to DragonflyDB. The post covers workload benchmarking, dual-read parity validation, zero-downtime rollout, and a shared-nothing failover mechanism based on cache-hit ratio analysis.

Agoda is conducting its 2026 AI Developer Report survey on how software engineers, engineering managers, and technology leaders across Southeast Asia and India are adopting agentic AI. The anonymous 15–20-minute survey will inform a free report on developers’ experiences, challenges, and expectations for autonomous AI in software engineering.

Agoda describes replacing direct Consul UI edits with a GitOps workflow for managing more than 34,000 configuration values across four data centers. The system adds JSON Schema validation, code review, ownership controls, audit trails, automated synchronization, and safer rollback, eliminating configuration-related outages after migration.

Agoda announces that its previously private bug bounty program is now public on HackerOne, covering core web services, APIs, and its mobile app. The post outlines eligible vulnerability classes, bounty tiers, reporting requirements, scope, and response-time commitments for security researchers.

Agoda describes a Workflow Simulator that tests flight supplier connectors through realistic, stateful booking flows without requiring a full end-to-end stack. The approach models workflows as DAGs, records interactions, and combines endpoint and cross-workflow assertions to detect mapping and integration defects earlier.

Agoda explains how it uses design tokens, layered themes, and shared React components to support multiple white-label travel brands from one codebase. The post covers theme package architecture, runtime CSS-variable injection, and Storybook-based tools for testing and editing brand-specific tokens.

Agoda engineers used continuous profiling and heap dumps to diagnose a Kotlin service suffering from unnecessary deserialization, three million queued coroutines, and CPU-bound serialization on IO threads. The post explains the fixes—including cache warming, bounded background work, and an upstream Koin upgrade—that reduced resource usage by approximately fivefold.

Agoda describes a multimodal content system that links hotel images, review snippets, facility data, and sentiment through a shared topic taxonomy. The post details its PySpark and Kubeflow pipelines, deduplication and ranking strategies, Couchbase serving model, and A/B-tested impact at large scale.