Browse
Log in / Register
Share
On May 9, 2026, Zoho founder Sridhar Vembu used X to deliver a blunt message to software engineers: use AI, but don't let it think for you. His comment came in response to a viral post from an engineer who described a company where Anthropic's Claude Code now produces nearly every piece of software work, leaving humans to do little more than press Enter. Vembu's warning has reignited a debate about AI coding assistants and what they are doing to engineering culture.
The engineer, whose employer was not named, said specifications, code, tests, product requirement documents (PRDs), tickets, ticket resolutions, and reports were all generated by Claude Code. "Nobody knows anything here. The specs, code, tests, PRDs, tickets, resolution of those tickets, reports, etc., everything is made by Claude Code. Nobody on my team likes this. They are being forced to ship as much as they can," the post read.
Vembu quoted that post and responded: "This is sad. I tell our engineers to use AI but never cede our understanding to AI." He went further, likening the industry's AI adoption to driving a fancy new car without knowing how to drive it. "As an industry, we are driving too fast in a fancy new car we barely understand how to drive, inviting disaster," he wrote.
Claude Code is Anthropic's agentic coding tool, launched widely in 2025. Unlike earlier code completion tools, it can handle multi-step tasks: writing a test, finding a bug, updating a ticket, or even drafting a full feature. That makes it powerful. It also makes it easy for an organization to hand over the entire software development lifecycle without anyone noticing exactly when human judgment slipped away.
The exchange is not just another CEO opinion. It touches on a growing frustration inside many software teams: AI is being used to increase speed, but the people doing the work are losing the opportunity to understand the system they are building. The original poster said employees were working 12 to 13 hours a day, not because they were writing more code, but because they were managing an AI assembly line. "People are working 12 to 13 hours a day just to press enter," the engineer said.
The engineer described the work as "soul-sucking" and said there was no sense of victory. "Nobody is resolving bugs. In reality, nobody is thinking anymore. Everything is done by LLMs. It is so soul-sucking. I would not mind it, to be honest, if we were at least given the time to check out the code and see what is going where. But no, the goal is to just ship. No matter what happens," the post said.
Vembu's reply stands out because he is not opposed to AI. He has been a proponent of AI tools in software engineering, and Zoho has integrated AI across its own products. The tension in his message is about ownership. As he put it, engineers should not "cede" understanding to a model. That means retaining the ability to explain why code works, not just what it does.
The debate around AI coding assistants has shifted in recent months. Tools such as Claude Code, OpenAI Codex, and GitHub Copilot have moved from novelty to default workflow in many companies. Executives see clear productivity gains. But the engineer's post is a reminder that productivity gains can come with hidden costs: worse code comprehension, fewer opportunities for junior engineers to learn, and a workforce that feels more like a supervisor of machines than a builder of software.
From a newsroom perspective, the most important lesson in Vembu's warning is about accountability. When no human understands the code, no human can be held responsible for its defects. That is a serious problem for security and reliability. A model-generated patch may pass tests, but it can also introduce subtle vulnerabilities that only a human with context would catch. Vembu's "never cede understanding" is not an anti-AI slogan; it is a way of saying that human judgment must stay in the loop.
The engineer's experience also shows how management incentives can undermine that idea. When the only measurable goal is shipping, there is little reward for spending time reviewing AI output or exploring how the parts fit together. The result is a team that produces a lot of code but understands less of it. If AI tools keep improving, the gap between output and understanding will only widen — unless leaders deliberately protect time for review.
There is another dimension here that often gets lost: the effect on junior developers. If all code is written by AI, and senior engineers are too busy shipping to explain it, then the next generation of engineers will never learn how to write code in the first place. Vembu's message, in that sense, is not just about quality today. It is about preserving the skills the industry will need tomorrow.
Zoho has a unique position in this conversation. The company builds enterprise software and operates with a long-term, private-company mindset. Vembu has spoken publicly about the value of deep work and craftsmanship. His advice to Zoho engineers — use AI, but keep your understanding — is consistent with that culture. For him, AI is a force multiplier, not a substitute for thinking.
Of course, not every company will follow that path. Some will use AI to aggressively reduce headcount. Some will turn engineers into prompters. The market may even reward that for a while. But the engineer's post suggests this approach is already burning people out. And when the industry inevitably finds bugs that no one can explain, Vembu's words may sound prophetic.
What remains to be seen is whether engineering leaders listen. Building a culture where people are expected to understand AI-generated code takes time and money. It may slow delivery in the short run. But as Vembu implies, the alternative is moving so fast that something breaks — in the code, in the team, or in the company as a whole.
For now, his message to engineers is simple. Use AI. Let it draft code, write tests, and resolve issues. But never hand over the part of the job that matters most: understanding.
