Sridhar Vembu, Chief Scientist and Co-founder, Zoho has shared a candid, experience-led perspective on the evolving role of AI in software development, drawing from an internal deep dive into AI-generated code at the company. In a recent post on X (formerly Twitter), Vembu revealed that Zoho’s engineering leadership spent hours reviewing C++ code generated by Anthropic’s Claude Opus 4.5 model during a technical town hall. The exercise, he noted, provided valuable clarity on what today’s AI models do exceptionally well, and where they still fall short.
Also read: Zoho Founder Sridhar Vembu Issues Clarification on Arattai
“Yesterday we had a tech town hall in Zoho where we did a code review of the C++ code generated by the Claude Opus 4.5 model. It went on for hours late evening,” Vembu said. According to him, AI excels at producing what engineers often describe as “glue code”, which is the connective tissue that moves data between systems, reshapes it, and ensures interoperability. While not conceptually complex, this kind of code is typically tedious and time-consuming for human developers.
Also read: Indians on H1-B visa in America Must Come Back Home: Sridhar Vembu
“I now have a much clearer understanding of what these models do well: they are able to stitch together systems well. There is often a lot of such ‘glue code’ in these systems and that is not very complicated but it is very tedious,” he explained. Vembu also pointed out that modern AI models appear to have effectively internalised vast swathes of open-source code, enabling them to recall familiar patterns and assemble solutions rapidly, though not without the occasional risk of hallucination. Crucially, he stressed that human expertise remains indispensable.
“Our senior engineer had guided (‘orchestrated’ is the right word) this process. When the AI was stuck he helped ‘unstuck’ it. This was a very vital contribution and without his experienced guidance, the AI output would not be useful,” he noted. The Zoho team reportedly reviewed multiple C++ files running into thousands of lines, searching for truly complex logic. What they found was revealing. “Most of it was straightforward glue code and only a tiny part of it was complex,” Vembu said, adding that the AI-generated output also appeared “needlessly verbose,” a point he plans to study further.
“On the whole, I am both impressed and not super awed. I believe we can do better,” he remarked, striking a balanced tone amid widespread hype around AI coding assistants.
In a related reflection, Vembu shared a more personal anecdote that highlights AI’s growing impact when paired with skilled engineers. He recounted how an experienced Zoho R&D engineer independently built an advanced assembly and machine-code security tool in just one month, something that would traditionally take a small team nearly a year. “He told me he found the Opus 4.5 AI model to be a game changer. Until that model, he was not all that enthusiastic about AI generated code but now he has revised his opinion,” Vembu said.
For Zoho, these experiments reflect a broader philosophy of learning through exploration. “This is how we learn in Zoho, we let smart people experiment and find new pathways,” he added. Yet, Vembu did not downplay the disruption ahead. Drawing a vivid analogy, he compared AI-powered development to the arrival of machine looms challenging traditional handloom weavers. “Powerful machine looms have arrived for software development, and the implications are enormous,” he wrote. As Zoho navigates this transition, Vembu acknowledged the pressure it places on leadership. “At Zoho, we have our work cut out, and as Chief Scientist, I am under pressure. Wish us luck.”
