Zoho's Sridhar Vembu Just Unveiled a Jaw-Dropping Truth About AI Code—You Won't Believe What He Discovered!

In a recent tech town hall at Zoho, founder Sridhar Vembu shared his insights on the use of large language models (LLMs) in coding, specifically highlighting a review of C++ code generated by Anthropic’s Claude Opus 4.5 model. The session, which lasted several hours, provided Vembu with a clearer understanding of the capabilities and limitations of AI in programming tasks. While he noted significant advancements in AI's coding abilities over the past two years, he emphasized that human intervention remains crucial for handling more complex coding scenarios.

Vembu articulated that during the review, AI excelled in producing what he termed “glue code.” This refers to the foundational code that integrates different systems, reshaping and transferring data effectively from one system to another. He explained, “I now have a much clearer understanding of what these models do well: they are able to stitch together systems well, taking data from one system, reshape it and pass it to another system. There is often a lot of such 'glue code' in these systems and that is not very complicated but it is very tedious.” His comments reflect a broader recognition in the tech community of AI's strengths in automating mundane coding tasks.

Vembu also pointed out some of AI's shortcomings, particularly its tendency to create overly verbose outputs and its reliance on patterns learned from existing open-source code. He stated, “In general, AI models have 'memorized' all the open source too and they are able to recall patterns from them (with some possibility of hallucination). They are also able to stitch various open source pieces together well.” This acknowledgement emphasizes an ongoing concern within the industry about AI's capability to generate novel code versus merely assembling existing patterns.

The Role of Human Oversight

One of the key takeaways from Vembu's review was the vital role that human oversight plays in the coding process. He specifically mentioned that during the review, a senior engineer was instrumental in guiding the AI's output. “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,” Vembu remarked. His observation reinforces the idea that while AI can assist in programming, it is not yet a complete replacement for human expertise.

In examining several C++ files, Vembu noted that most of the code generated was straightforward glue code, with only a small fraction considered complex. He further speculated that the AI-generated code often leaned towards being needlessly verbose: “I suspect that the AI generated code tended to be needlessly verbose but I have to study it more to be sure. On the whole, I am both impressed and not super awed. I believe we can do better.”

This sentiment is echoed across the tech landscape, where many developers are optimistic about the potential of AI, yet cautious about its current capabilities. As companies increasingly adopt AI tools to streamline workflows and enhance productivity, the balance between automation and human oversight will be a critical factor in determining the success of these technologies.

The advancements in AI-driven coding tools signal a significant shift in how software development may evolve in the coming years. As highlighted by Vembu, the journey towards fully autonomous coding is still in its early stages, necessitating a collaborative approach that combines the strengths of both AI and human expertise. For American readers, this evolving narrative underscores the importance of staying informed about how technology impacts various job sectors, particularly in software development, where the integration of AI could redefine roles and responsibilities.