Earlier this month, Agoda officially opened its new 26,000 square metre APAC Tech Hub in Bangkok, spread across seven floors in One Bangkok. Around 4,000 of Agoda’s more than 7,000 employees will be based there, including about 1,500 of its 2,000-strong tech team.
In this first of a two-part feature, Frontier Enterprise spoke with Agoda CTO Idan Zalzberg during an office tour in Bangkok. Zalzberg discussed the strategy behind the office move, how the company approaches innovation, and how its IT infrastructure differs from when he joined over a decade ago.
By the numbers
Previously offering only hotel bookings on its platform, Agoda has since expanded its listings to include flights, car rentals, and activities.
During a media presentation, Agoda CTO Idan Zalzberg said the company lists around 6 million properties globally and sees about 90 million monthly visitors. He added that Agoda was the most downloaded app in 2025 in the OTA category, and that its flights business supports 130,000 routes across multiple languages and markets.
In terms of scale, Agoda processes 4.33 trillion messages every day across its messaging clusters, with about 50 million messages per second at peak. Meanwhile, the company is also calculating 30 million prices every second for its customers.
Zalzberg said Agoda writes about 580 terabytes of data a day and reads about 14 petabytes daily.
“For generative AI, we are processing about half a million tokens per second on average, reaching over 1 million tokens per second at peak across multiple providers and a load-balanced set of computers,” he added.
In the zone
Although headquartered in Singapore, Agoda was founded in Phuket in 2005, which is why opening its largest office space to date in Thailand aligns with its origins. An integral part of Agoda’s One Bangkok office is the Network Operations Centre (NOC) Zone, which provides a dedicated space to monitor systems and manage operations.
“You have to make sure that every aspect is being looked at, and make sure that everything is healthy. A great majority of this comes through automation, and these dashboards use AI to indicate when something is wrong,” Zalzberg explained as he toured tech journalists across the NOC Zone.
Despite the level of automation, human oversight remains necessary, Zalzberg said.
“We do want to have that human aspect here. We want to have that human common sense and judgment to say whether something feels wrong, and also to manage the actual operations,” he said.
In the NOC Zone, Agoda uses generative AI mainly for cause analysis and alert validation, two interconnected processes which previously took up a significant amount of the IT team’s time.
“We get a lot of alerts, and the problem is, how do you know which ones are problematic? Sometimes, when something actually goes wrong, there would be a number of signals coming up. But in order to look at that together, see the full picture, and understand what happened, this is where generative AI can come in,” Zalzberg noted.
He said Agoda uses an internally built SRE-bot to analyse alerts and production changes, providing context to humans during validation.
“What’s useful about AI is that it can process large volumes of text, so it can look at error logs and the code changes that were deployed, and it does that in seconds. It can review what was deployed in a recent change, examine the code changes, and determine whether they align with the issue it has detected,” Zalzberg said.
Employee support centre
Within Agoda’s seven-floor office in One Bangkok, there is also a service centre called Employee Connect, where Agodans can go to resolve issues, from laptop replacements to visa matters. Zalzberg said the mechanism powering Employee Connect was also internally built.
In front of the service centre are several touchscreens where employees can choose which service they want to access. They then receive a queue number and wait for their turn to speak to an expert behind the counter, similar to a bank or service centre.
Zalzberg said Employee Connect was well received by Agodans because it simplified how they resolve their needs.
“If you need to have your laptop fixed, or software updated, you no longer need to file a ticket then wait for somebody to respond,” he said.
Massive leap
Zalzberg, who has been with Agoda since 2014, has witnessed how the company’s IT stack has evolved. “We just started with VMs (virtual machines) at the time,” he recalled.
Before VMs, it was bare metal, Zalzberg said. During the early days of VMs, provisioning remained a time-consuming, often siloed process.
“You have to fill up a form to ask for a VM. You have to wait for someone to look at it, have someone else do it for you, and then after a few days, maybe you get access to that machine. You install stuff, nobody knows what you’re doing there, and you start it. You have your own deployment and your own process,” Zalzberg noted.
Today, provisioning is largely self-service, as teams are expected to operate without waiting on others if the company wants to maintain a culture of ownership, Zalzberg said.
“Today, everything is visible and governed, so you do it yourself, but you do it in a way where you declare exactly what you are deploying. You have the controls, but then we get the visibility, so everything is visible out of the box — automated alerts, automated AI, processes that look at everything, we make sure the governance and the compliance is there,” he explained.
Zalzberg added that Agoda’s internal tools are cloud-compatible, giving the company flexibility in how it deploys workloads.
“If in the future, for example, pricing is more favourable, we can choose to move to the cloud. We developed our technology in accordance with cloud protocols such as S3 or Kubernetes, so what we run today can also run on the cloud. We sometimes run tests on the cloud, or use it where compliance requires a small cluster in a specific country. Because we follow standard protocols, we’re not locked into a system that’s disconnected from the wider ecosystem. It’s very easy for us to run on-premises or in the cloud,” Zalzberg said.
(to be continued)
