Zoho co-founder Sridhar Vembu offers measured take on AI-written code, thinks the bar can be...

Sridhar Vembu

• Zoho's Vembu says AI speeds up routine coding but needs human guidance.
• Engineers must refine and simplify AI-generated code before deployment.
• Human expertise essential for quality, maintainability in complex coding tasks.

Did our AI summary help?

Sridhar Vembu, co-founder of Zoho, has shared a nuanced view on the current state of AI-generated software code, saying he is impressed by recent progress but not “super awed.” His comments come after an internal review exercise at Zoho that closely examined large volumes of code written with the help of an AI system.

Inside Zoho’s AI code review

Vembu said the discussion took place during a technical town hall where senior engineers reviewed several thousand lines of C++ code generated by an AI model. The session ran for hours and focused on understanding not just whether the code worked, but how it was structured and how much human effort was still required to make it usable in real-world systems.

According to Vembu, much of the AI-generated output fell into the category of routine “glue code.” This included tasks such as integrating different software components, handling data flow between modules, and implementing standard patterns that experienced developers are already familiar with. In these areas, the AI proved useful by speeding up work that is often repetitive and time-consuming.

Human guidance remains central

However, Vembu stressed that the AI did not operate independently. A senior Zoho engineer was deeply involved in guiding the system, correcting its direction when needed, and deciding what should or should not be accepted. Without this orchestration, the resulting code would not have met production standards.

He also noted that only a small portion of the reviewed code involved genuinely complex logic. For such sections, human judgment was critical to evaluate correctness, performance implications, and long-term maintainability—areas where AI still struggles to make reliable decisions on its own.

Code quality and efficiency questions

Vembu observed that the AI-generated code tended to be verbose in places. While verbosity is not always a flaw, it can make systems harder to maintain and understand over time. This highlighted the need for experienced engineers to refine, simplify, and sometimes rewrite parts of the output before deployment.

A balanced view of AI’s role

Summing up his experience, Vembu said the exercise left him both impressed and unconvinced that AI has reached a ceiling. His takeaway was that AI is a powerful productivity tool when paired with strong engineering leadership, but it is not yet a replacement for deep technical expertise.

“I believe we can do better,” he implied, suggesting that future improvements—both in AI systems and in how teams use them—could push the bar higher for software development productivity and quality.