Software developers are turning to artificial intelligence (AI) for clear productivity gains, but there are gaps that need to be plugged and some limits to be put in place.
These will ensure developers still are able to make good decisions and remain fully accountable for their work.
Job candidates, for one, should be able to demonstrate their ability to code without the use of any AI, according to Agoda CTO Idan Zalzberg, who spoke with FutureCIO in a video call.
Amongst the qualities Zalzberg looks for in candidates is their coding prowess, sans AI.
“I want to assess their ability based on what they know…I’m not ready just yet to say I don’t care if you can code,” said Zalzberg, who joined the travel tech company in 2014 and was appointed CTO in 2022, leading some 2,000 engineers. Agoda has a total headcount of about 8,000 and operates its tech infrastructure across four data centres, including Singapore and Hong Kong in Asia, Amsterdam in Europe, and Ashburn in the US.
Zalzberg, though, encourages the use of AI in his team of developers, who he believes should have access to the necessary tools and be able to experience the impact AI has had on the industry.
“There have been big changes in software development over the past decade, not just the past year, but this feels like the first time we’re not coding in a very structured way,” he said.
Moving time to where it matters
One significant difference is that developers are spending less time writing syntax, and more time conversing with the agent writing the code, he noted.
Their time now has transitioned from having to write the text to making decisions, for instance, on how to address a problem or to implement something, he said.
“This is more material than writing the right syntax. That’s how I think the work has shifted,” Zalzberg said, adding that developers are no longer limited by how fast they can type and remember syntax. “It allows people to make good decisions. Now your speed is limited by how fast you can think and come up with the right architecture, the right solution."
In fact, 95% of developers in this region already are using AI weekly, according to a study released by Agoda in November 2025. Another 56% say they always keep an AI assistant open.
The online survey was conducted by Macramé Consulting last August to September, which polled more than 600 software developers across India and seven Southeast Asian markets, including Singapore, Thailand, Indonesia, and the Philippines.
Some 80% of respondents point to speed and automation as the key drivers of their AI adoption, with 37% estimating they save four to six hours per week from their AI use.
Some 72% see productivity gains and better code outcomes from their AI use.
However, while 94% depend on AI for code generation, just 22% of developers use the technology to solve unfamiliar problems.
Some 43% rate AI’s performance as similar to a mid-level engineer, with 79% highlighting inconsistent or unreliable outputs as the main barrier to wider use of the technology.
In addition, 67% review AI-generated code before merging and 70% say they routinely rework outputs “to ensure correctness”, the Agoda study noted.
There currently still are gaps, said Zalzberg, adding that he believes these will be plugged as AI continues to advance.
For instance, the technology currently is less reliable in making the right decisions with large software projects involving millions of lines in codes, or where many external systems have to be integrated in the project.
Supervision and accountability required
AI agents, for one, need babysitting and human guidance on what they should and should not be doing.
And while AI appears most mature or stable in the coding phase of the software development lifecycle, it still has some ways to go in code review and monitoring.
These issues will be addressed as AI agents and models continue to train and learn, said Zalzberg, who was previously Agoda’s chief data officer, where he was responsible for managing Agoda’s data and security solutions.
Simply pushing more data to train AI models isn’t enough, he said, as he stressed the importance of context engineering.
“AI models are smart but not infinitely smart. If you throw all the data you have, it will be confused just like humans,” he noted. “It won’t pick up all the subtleties. It’s about choosing which data, at the right time, and the right LLM (large language model) it can [best] work with.”
For now, companies need the experience and engineers who can establish the context and come up with viable solutions. For instance, AI agents will not realise changing the code in a software used in one area of the procurement chain, may impact another area, such as payments.
And, ultimately, developers--not the AI agents--still are accountable for the codes and their efficacy, he said.
On its part, Agoda works to ensure its developers can leverage the AI tools they need to be the most productive and to remain competitive in the market, according to Zalzberg.
The digital travel company also has safeguards and processes in place to assess and mitigate potential risks, he said. For instance, its generative AI (GenAI) use cases are mostly low risk.
Agoda runs some 200 initiatives that leverage GenAI, including in customer support and user interface, so users have better information before making their bookings.
Its employees also maintain full accountability and cannot attribute any error to AI, Zalzberg added.
This motivates them to better manage their AI use, so they are not simply offloading responsibility, he said.
The most common pitfall companies make is assuming AI already is working perfectly, he further noted.
While the technology may work most of the time for most use cases, it currently still is susceptible to hallucinations and mistakes, he said.
Organisations need to be ready to adapt to the reality that AI does not function at the same level of trust and consistency as traditional, standard software, Zalzberg said.
This underscores the importance of implementing the necessary processes in their AI use, including evaluation, measurement, and feedback loop, he said.
According to the Agoda study, one in four teams use AI under official guidelines from their organisation.
Most developers are self-taught in their adoption, including 71% who learn through tutorials, side projects, or online communities.
Some 28% underwent training via their organisation, with developers in Singapore almost twice as likely as those in Vietnam to have formal AI training programs, the study found.
