Sridhar Vembu AI Coding Warning: Why Zoho's Founder Says Developers Should Not Give Up Their Understanding to AI
Artificial intelligence is changing software development at a speed that would have seemed unrealistic only a few years ago.
AI tools can now generate code, write tests, prepare technical documents, analyse errors and help developers complete software tasks in a fraction of the traditional time.
But Zoho co-founder and chief scientist Sridhar Vembu has raised a different question:
What happens if developers become so dependent on AI that they no longer understand the software they are building?
That question is behind Vembu's latest warning about AI-assisted coding, which has attracted significant attention among Indian technology professionals and developers.
What Did Sridhar Vembu Say About AI Coding?
Vembu recently responded to a software engineer's post describing an extremely AI-dependent development environment.
According to the engineer's account, an AI coding tool was being used across almost the entire software development workflow — from specifications and code to tests, product requirements, tickets, ticket resolution and reports.
The engineer claimed that employees were under pressure to deliver as much software as possible while spending much of their time directing the AI.
Vembu reacted by saying that engineers should use AI, but should not surrender their understanding of the work to it.
His concern was therefore not simply about AI-generated code.
It was about what happens when the human engineer stops being able to explain, inspect, debug or take responsibility for the software being produced.
Vembu Is Not Saying Developers Should Stop Using AI
This is one of the most important parts of the story.
The headline can easily be interpreted as:
"Sridhar Vembu is against AI coding."
That isn't an accurate description of his position.
Vembu has said that Zoho engineers use AI in programming.
His concern is about how AI is used, rather than whether it should be used at all.
There is a major difference between:
Developer + AI
and
AI + human approval button
In the first model, AI helps the engineer work faster.
In the second model, the engineer may simply accept whatever the AI produces without understanding the underlying system.
Vembu's warning is focused on the second scenario.
Why Is AI Coding Creating This Debate?
Modern AI coding tools are becoming much more capable.
They are no longer limited to completing a few lines of code.
Depending on the tool and workflow, AI can help with:
Generating software code
Explaining existing code
Finding potential bugs
Creating tests
Refactoring programs
Writing documentation
Working with repositories
Creating development tasks
Debugging problems
Building prototypes
Automating repetitive engineering work
This creates an obvious productivity opportunity.
A developer can potentially complete work much faster.
But faster production creates another problem:
Can humans review everything that AI produces at the same speed?
That is where the debate becomes complicated.
The Real Risk May Not Be AI-Generated Code
AI-generated code can be wrong.
But developers have always produced buggy code.
The bigger concern is what happens when people lose the ability to recognise that something is wrong.
Imagine a developer receives thousands of lines of AI-generated code.
The software passes automated tests.
The application appears to work.
But a hidden security vulnerability exists somewhere inside the system.
If the developer understands the architecture, they have a better chance of identifying the problem.
If they don't understand the system and simply trust the AI output, finding that vulnerability can become much harder.
This is why technical understanding still matters even when AI writes a large portion of the code.
Why India's Developers Are Paying Attention
This story has particular relevance in India because software engineering is a major part of the country's technology economy.
Millions of professionals work across:
Software development
IT services
Application maintenance
Cloud engineering
Testing
Technical support
Data engineering
Cybersecurity
Consulting
AI-assisted development could change how many of these roles operate.
For Indian developers, the question isn't simply:
"Will AI write code?"
AI already can.
The more important question is:
"What skills will remain valuable when AI can write much of the code?"
That is where Vembu's argument becomes relevant.
Coding Skills May Become Less Valuable by Themselves
Knowing how to write syntax correctly has traditionally been an important part of software development.
AI is changing that equation.
If a developer can describe what they want and an AI system can generate the implementation, typing code manually may become less important for some tasks.
But other skills could become more important.
For example:
Understanding the business problem
A developer needs to understand what the software is actually supposed to accomplish.
Understanding system architecture
Knowing how different services, databases and applications interact remains important.
Security awareness
AI-generated code can still introduce security problems.
Debugging
When something goes wrong, someone needs to understand why.
Technical judgement
Developers need to decide whether an AI-generated solution is appropriate.
Domain expertise
Knowledge of healthcare, finance, telecommunications, manufacturing or other industries can help developers make better technical decisions.
AI can generate code.
It doesn't automatically understand the real-world consequences of every decision.
Vembu's Bigger Message: Don't Outsource Your Brain
The most interesting part of this discussion isn't actually about coding.
It is about human responsibility.
AI can become extremely useful when it acts as an assistant.
But problems can emerge when organisations treat AI as an unquestionable authority.
A developer should be able to ask:
Why was this approach selected?
What does this code actually do?
What assumptions does it make?
What happens if the input changes?
Could this create a security vulnerability?
How will it behave at scale?
What happens when the AI-generated solution fails?
If nobody on the team can answer those questions, the organisation may have increased its coding speed while reducing its technical control.
The "Faster Coding" Trap
There is a natural temptation for companies to measure AI adoption through productivity numbers.
For example:
More code → more features → faster releases
But software quality is not determined by the number of lines of code produced.
In fact, good engineering often involves writing less code, simplifying systems and preventing unnecessary complexity.
AI can make it extremely easy to generate software.
That doesn't necessarily make it easy to maintain that software.
A company could therefore find itself with a large amount of AI-generated code that nobody fully understands.
That could create a maintenance problem several years later.
What Happens to Junior Developers?
This may be one of the most important questions raised by the AI coding revolution.
Junior developers traditionally learn by doing.
They write code.
They make mistakes.
They investigate bugs.
They read documentation.
They ask senior engineers questions.
They gradually develop an understanding of how software systems work.
If AI performs most of those tasks for them, juniors may become productive very quickly.
But there is a potential downside.
They could become dependent on AI before developing the underlying skills needed to judge its output.
That creates a strange situation:
A junior developer may be able to produce more software while understanding less of it.
This is one reason companies may need to rethink how developers learn in an AI-assisted workplace.
AI Could Make Good Developers More Productive
There is another side to the story.
AI coding tools can be extremely useful.
A developer who already understands a system can use AI to:
Automate repetitive tasks
Generate boilerplate
Explore possible solutions
Create initial test cases
Explain unfamiliar code
Speed up debugging
Build prototypes
Experiment with alternative approaches
In this situation, AI doesn't replace understanding.
It amplifies understanding.
This may be closer to the model Vembu is advocating.
The engineer remains responsible while AI acts as a powerful development assistant.
What Does This Mean for Software Jobs?
The answer isn't as simple as saying:
"AI will replace programmers."
Some programming tasks are likely to become increasingly automated.
Other responsibilities may become more important.
Developers could increasingly spend their time on:
Architecture
System design
Product decisions
Security
Reliability
AI supervision
Integration
Testing strategy
Domain-specific problems
Reviewing AI-generated work
In other words, the definition of a software engineer may gradually change.
The person who is valuable may not be the person who can type code the fastest.
It may be the person who can understand a complex problem and use AI effectively to solve it.
Why Claude Code Is Part of the Discussion
The recent debate was triggered by an account describing extensive use of Claude Code, Anthropic's AI coding tool.
The important point isn't that Claude Code is uniquely responsible for the concern.
Other AI coding systems can also perform increasingly sophisticated development tasks.
The broader issue is the transition from:
AI suggests code
to:
AI participates in large parts of the software development process.
That transition changes the role of the developer.
The more responsibility given to an AI system, the more important human review becomes.
Could AI-Generated Software Become a Security Problem?
Potentially, yes.
The concern is not that AI-generated code is automatically insecure.
The problem is that developers may assume that code is safe simply because an AI produced it or because automated tests passed.
Security requires more than functional testing.
A system can behave correctly during normal testing and still contain vulnerabilities.
This is particularly important for software handling:
Customer data
Financial information
Passwords
Authentication
Personal information
Healthcare information
Business systems
The more powerful AI coding becomes, the more important independent checks and human accountability may become.
What Is Zoho's Approach?
Vembu has described Zoho as using AI extensively in programming while keeping human developers responsible for the software they release.
The philosophy he has discussed can be summarised as:
Productivity with safety.
That is an important distinction.
The goal isn't necessarily to slow down AI adoption.
The goal is to make sure that faster development doesn't remove the safeguards required to understand and maintain the resulting software.
What Should Developers Learn Now?
The AI coding revolution doesn't mean developers should stop learning programming.
It may mean they need to learn programming differently.
Instead of focusing only on memorising syntax, developers should build strong fundamentals in:
Computer science
Understand how software actually works.
System design
Learn how applications interact and scale.
Databases
Understand how information is stored and retrieved.
Networking
Know how systems communicate.
Security
Understand common vulnerabilities and defensive practices.
Debugging
Learn how to find the real cause of a problem.
Testing
Understand what tests can and cannot prove.
AI tools
Learn how to use AI efficiently while reviewing its output.
Domain knowledge
Understand the industry and customers for whom the software is being built.
These skills can make AI more useful rather than making the developer more dependent on it.
The New Developer May Be an AI Supervisor
One possible future is that developers will write less code manually.
Instead, they may spend more time:
Defining the problem
Designing the architecture
Giving AI precise instructions
Reviewing generated solutions
Testing the system
Finding weaknesses
Making architectural decisions
Taking responsibility for the final product
That doesn't necessarily make software engineering less important.
It changes what engineering work looks like.
The difficult skill may shift from writing every line to knowing which lines should exist in the first place.
Why Sridhar Vembu's Warning Matters
The most important takeaway from Vembu's comments isn't:
"Don't use AI."
It is:
"Don't stop understanding what AI is doing."
That distinction is increasingly important as AI coding systems become capable of handling larger portions of software development.
A developer who uses AI while maintaining strong technical knowledge may become significantly more productive.
A developer who uses AI without understanding its output could become dependent on a system they cannot effectively troubleshoot when something goes wrong.
Final Takeaway
Sridhar Vembu's latest AI coding warning comes at a time when software development is undergoing a major transformation.
AI can already write substantial amounts of code and assist with many stages of development.
The challenge now is not simply figuring out how much AI can do.
It is figuring out how much responsibility humans should give it.
For developers, particularly those entering the profession today, the lesson may be straightforward:
Learn to use AI, but don't let AI become the only thing you know how to use.
The most valuable developer of the future may not be the person who writes the most code manually.
It could be the person who understands the problem deeply, knows when AI is useful, recognises when its output is wrong and remains accountable for what ultimately reaches the user.
Frequently Asked Questions
What did Sridhar Vembu say about AI coding?
Sridhar Vembu said engineers should use AI coding tools but should not give up their understanding of the software they are building.
Is Sridhar Vembu against AI coding tools?
No. His comments indicate that he supports using AI as a productivity tool while emphasising human understanding, technical judgement and responsibility.
Why is Sridhar Vembu's AI warning trending in India?
His comments are attracting attention because they relate directly to software engineering, AI coding tools and the future of technology jobs in a country with a large IT and software workforce.
What is Claude Code?
Claude Code is an AI-assisted software development tool from Anthropic that can help developers perform coding and other software-engineering tasks.
Will AI replace software developers?
AI is likely to automate some software-development tasks, but the current debate does not establish that software developers as a profession will disappear. The role may evolve toward architecture, review, security, system design and AI-assisted development.
Should developers still learn coding?
Yes. Strong programming and computer-science fundamentals can help developers understand, evaluate, debug and safely use AI-generated software.
What is the biggest concern with AI coding?
One concern is that developers could become dependent on AI-generated output without developing enough understanding to review, maintain or secure the resulting software.
What is the main lesson from Sridhar Vembu's warning?
The central idea is that AI should increase human capability without replacing human understanding and responsibility.