At a tech townhall inside global enterprise software company Zoho on Friday, its founder and chief scientist Sridhar Vembu emphasised that while artificial intelligence (AI) is useful for coding, it cannot replace software engineers who “orchestrate” AI to produce desirable outcomes.

As AI increasingly acts as a productivity booster for coding, Vembu said a senior Zoho engineer actively “orchestrated” the model through guided prompts, intervened when it got stuck, and applied human judgement and experience to refine the output.

In a post on X, Vembu revealed that Zoho engineers recently reviewed C++ code generated by Anthropic’s Claude Opus 4.5 model.


C++ is a widely used programming language known for building high-performance software, such as operating systems, databases, and large enterprise applications.

Opus 4.5, unveiled in November last year by Anthropic, the Amazon- and Alphabet-backed company, was ranked among the most powerful models in the Claude family, offering deep reasoning and memory, coding, and a versatile performance across a range of computer applications, including financial tasks such as modelling and forecasting.

According to Vembu, the review showed that the generated code was fairly simple, with only a small fraction being truly complex. He added that the AI had also been needlessly verbose in its output.

“I suspect that the AI-generated code tends 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,” he added.

Vembu said AI works well at writing “glue code” that connects systems, moves data, and works with application programming interfaces (APIs), tasks that are often tedious for human developers.

An API is a set of rules that allows different software applications to communicate and exchange data with each other.

Additionally, he pointed out that AI demonstrates strong pattern recall, particularly for open-source code that it has effectively memorised. However, he cautioned that this also increases the risk of hallucinations.

The development is in stark contrast to Vembu’s remark last year, where he said AI would take over most coding tasks. He attributed this to the repetitive nature of programming work, which AI can efficiently handle.


However, he mentioned in the recent post that with the review of the C++ code developed by the AI model, he has a clearer picture. “I now have a much clearer understanding of what these models do well,” he said.

Also Read | Code it my way: Techies learn to 'vibe' with AI