From Traditional Coding to AI-Powered Development: A Google Engineer's Journey

A conversation with Priyank Bhadja, Senior Software Engineer at Google, who transitioned from Microsoft to Google during the AI revolution and now writes 60-70% of his code with AI assistance

August 2, 2025

12 min read

By Rachit Magon

From Traditional Coding to AI-Powered Development: A Google Engineer's Journey

Remember those days when writing code was everything? When we took pride in code that compiled on the first try and followed industry standards to the letter? Well, those days feel like ancient history now.

We sat down with Priyank Bhadja, a senior software engineer at Google who's lived through this transformation firsthand. From his days at BITS Pilani to Microsoft's Azure team, and now at Google's cutting-edge AI-integrated workflows, Priyank represents the evolution every developer is experiencing – or should be.

At Google, engineers are achieving 37% code completion rates with AI tools. But here's the thing: they're not coding less, they're coding differently. And honestly, they're solving problems they never could before.

Key Takeaways: Your Roadmap to AI-Native Development

Start Today (Next 7 Days):

• Set up Gemini CLI with a project-specific .gemini.md file for coding standards
• Begin using AI for research and quick prototyping instead of just Stack Overflow
• Practice breaking complex problems into smaller, AI-manageable chunks

Build Skills (Next 3 Months):

• Master prompt engineering for your specific domain and use cases
• Learn to review AI-generated code for security, logic, and integration issues
• Focus on "what and why" thinking instead of just "how to code"

Future-Proof (Next 1-2 Years):

• Develop AI-native architecture design skills
• Build deep domain expertise and critical thinking capabilities
• Learn advanced AI interactions: model tuning, complex workflows, AI ethics

Q: Can you take us through your journey from BITS Pilani to Microsoft and then to Google?

Priyank: Sure, thanks for having me on ChaiNet. I completed my masters in software systems from BITS Pilani in 2016. It was truly magical studying there, especially exploring parallel computing and cloud computing – remember, cloud was just a buzzword during our college time!

After that, I got placed at Microsoft through campus placement and worked on various Azure teams. Specifically, I worked on building the first-party service Azure Farm Beats, which is now known as Azure Data Manager for Agriculture. This was my first experience building a first-party cloud service from scratch.

We had seen issues with existing IaaS products and built a scalable Platform-as-a-Service solution that was much faster in execution. I worked on designing weather services, sensor services – it helped me understand how scalable cloud services are built.

This was the era when we used to write all code by ourselves, taking pride in writing code that compiles on the first go and follows industry standards.

🔥 ChaiNet's Hot Take: The pre-AI era of software development already feels nostalgic. Priyank's journey from Azure Farm Beats to Google's AI-integrated workflows mirrors the entire industry's transformation in just a few years.

Q: What happened when you entered Google and how did your relationship with AI tools actually develop?

Priyank: I switched to Google in 2022 and joined the GCP Storage, Backup and DR team. Having worked in Azure cloud, working and adopting development in GCP was exciting – I started understanding differences in the offerings of Azure and GCP as cloud platforms.

This is the era when ChatGPT and Bard were introduced. Later, Bard was renamed to Gemini in '23, and since then this AI era has started.

During college days, I don't think I was using any AI tools – it was kind of sci-fi movies only. In Microsoft also, most of the work I did was based on typical software engineering processes. There was no AI or LLMs during that time; the most resembling thing was machine learning classification and clustering.

I started using actual AI at Google when Bard was introduced. Initially, it was for basic know-how questions, replacing Stack Overflow to fix small issues. There were a lot of hallucinations, and using it for development wasn't that efficient.

After that came the era of Gemini, and all its LLM models were growing so fast that it seemed kind of magical to see how AI became a companion in my day-to-day life.

🔥 ChaiNet's Hot Take: The transition from "AI as sci-fi" to "AI as daily companion" happened in less than two years. Google's internal AI tools evolved rapidly from the early Bard days to today's sophisticated Gemini Code Assist.

Q: How did your mindset change once you started using AI as a coding partner?

Priyank: This is an interesting thing to understand – what AI can bring to our daily lives. I organically learned the AI tools, and what I saw when this mindset shift was happening is that I no longer need to worry about how to write the code. Rather, I focus on the what and why of the problems.

I just need to understand what are the weaknesses of AI. What I remember is that large context texts, overwhelming or bigger tasks, ambiguous tasks, or prompts prone to hallucination are going to give you inaccurate answers.

Learning how to break them into smaller problems which AI can code accurately, review the generated code thoroughly, and integrate all these pieces together to build the application – these are the crucial tasks.

🔥 ChaiNet's Hot Take: The shift from "how to code" to "what to build and why" represents the fundamental transformation in software engineering. It's not about writing less code; it's about thinking at a higher level.

Q: Do you think AI will replace developers? What's the reality inside Google?

Priyank: That's the most discussed question nowadays, both internally at Google and externally. At Google, where AI is deeply integrated into almost everything we do, the reality isn't about replacing developers. Rather, I would say it's transforming how we code and what it means to be a developer.

I'm coding differently rather than coding less. It's more about coding strategy. AI helps me do research, understand specific areas, and generate boilerplate code and repetitive code. So I get to focus more on core problems rather than coding from scratch.

You can basically focus on what's important and leave everything which is not important to AI.

As for junior engineers – I think there will be a shift in the paradigm. As long as we stay hungry, stay foolish, AI is not going to hamper their journey. But there will be a shift in the way juniors approach software engineering.

A lot of educational shift is required because if we go back to our BITS Pilani course and people are still being taught the same things, they'll lag behind really fast. We need to start teaching people how to take AI's help rather than fight it.

🔥 ChaiNet's Hot Take: Google's 37% code completion rate with AI tools supports this view. The company that's supposedly leading AI development sees it as transformation, not replacement. The key insight: educational institutions need to evolve rapidly.

Q: Can you give a specific example of how you use AI in your workflow at Google?

Priyank: I cannot disclose specific details because of NDA, but I can explain how I approach problems.

I first try to write the design myself in a document, then ask AI to review it from a high level and suggest changes based on best practices. Then I break down this design into individual components and do detailed research using Gemini – what's the best way to build the system? What database decisions should I take? What are the pros and cons of using one design over another?

AI may be helpful in some areas but may not understand the domain or the whole context around the feature and user's mindset. So I need to bring in that critical thinking while deciding the best design to choose.

Once the design is ready, I ask AI to write code for the technical components. I generate prompts, provide useful existing code as reference, and get components written by AI. Prompt engineering is a key factor here.

🔥 ChaiNet's Hot Take: This workflow showcases the hybrid approach: human strategic thinking combined with AI execution. The critical insight is that domain knowledge and context remain uniquely human.

Q: How does Google upskill its engineering teams to stay ahead in AI capabilities?

Priyank: As AI has been changing so fast, we need to keep ourselves ahead of AI. We need to make learning a constant part of our job – that's the first thing.

Apart from that, we have structured education systems where we get internal courses, workshops are handled, there are guides on AI tools and responsible use of them. These are useful resources that help Googlers internally advance their skills and use effective AI prompting.

But honestly, on-the-job practice gives you the best shot at learning new things. Get your hands dirty, do hands-on, learn new things – these are key factors that help us get ahead.

🔥 ChaiNet's Hot Take: Even at Google, there's no secret sauce. The fundamentals remain the same: continuous learning and hands-on practice. The difference is having structured resources and a culture that embraces experimentation.

Q: Walk us through a typical day when working on complex features with AI.

Priyank: A typical day starts with planning and designing the feature. It requires clarifying the problem statement first – understanding high-level goals, doing brainstorming, whiteboarding, and focusing on what needs to be built.

AI helps in various ways: summarizing existing documents on existing products, learning key design choices that similar products have made. This needs to be analyzed by humans to ensure the right learnings are carried forward.

Once design is finalized, we use AI to generate diagrams and define the skeleton of the solution. Then AI helps in execution – this is where I benefit from internally integrated Gemini Code Assist.

Gemini Code Assist is integrated with Google's internal code repository. It helps me write code through pair programming using AI. I need to clearly define tasks in technical terms and keep reviewing and refining AI-generated code, otherwise there can be hallucinations and missed edge cases.

Once code is generated, I test it thoroughly and ensure all edge cases are covered. AI helps write unit tests as well.

🔥 ChaiNet's Hot Take: The integration of AI tools with internal repositories gives Google engineers a significant advantage. The workflow shows AI as a collaborator throughout the entire development lifecycle, not just code generation.

Q: What percentage of your code is AI-written versus your own code?

Priyank: Given the nature of my work, which is more on exploring new areas and building prototypes to propose features, AI helps me write 60 to 70% of the code.

Sometimes AI may hallucinate while writing code and use client library functions that don't even exist. In such cases, it becomes very important to consult official documentation or provide inputs that help AI avoid hallucinations and guide AI in the right direction to write correct and complete code.

🔥 ChaiNet's Hot Take: 60-70% AI-generated code at Google represents the current state of AI-assisted development. The remaining 30-40% isn't just human code – it's human oversight, domain expertise, and critical thinking that makes the AI code work.

Q: How has AI changed your approach to system architecture?

Priyank: AI has helped and changed the way we approach architecture designs. It can help with generic and shallow solutions initially from a backend system or design perspective. However, it becomes a very good companion when it comes to researching different ways of building individual components.

AI can help evaluate your design critically to identify any missed nuances or best practices. However, when you're writing a whole integrated design, that still requires critical thinking from humans. In design cases, you still need human intuition and critical thinking.

Where AI excels: predefined tasks with clear definitions, like customer support bots with clear instructions. It has expertise in clustering, classification, and pattern recognition problems. It can help with repetitive tasks like writing boilerplate code.

AI has performed surprisingly well in one area – writing front-end code. You might have seen demos where you upload hand-drawn UI charts and AI converts them into UI.

🔥 ChaiNet's Hot Take: The insight about AI excelling at front-end development is particularly interesting. Visual-to-code translation represents one of AI's strongest current capabilities in software development.

Q: How have code reviews and team collaboration changed at Google?

Priyank: AI has shifted team collaboration. We have AI assistance handling much of the boilerplate code and routine code generation, so this frees up engineers for more complex problem-solving.

Code reviews haven't fundamentally changed in process – the same rigorous standards of human oversight from peers, code owners, and language experts are definitely needed. However, the focus of review is changing from catching simple syntax errors to validating logic from a security perspective.

You also need to ensure well-integrated code because with AI-generated components, you never know whether they'll integrate well. Multi-layer reviews from peers, AI agents, and internal AI tools help ensure that code changes meet required standards.

🔥 ChaiNet's Hot Take: The evolution of code reviews from syntax checking to logic validation represents a maturation of the development process. Human reviewers can focus on higher-level concerns while AI handles routine checks.

Q: What's your advice for developers whose CTO just said "start using AI"?

Priyank: Instead of just directly jumping on tools, it's better to have the right mindset and culture in place. We need to ensure that people are learning AI tools – we cannot assume that tools exist and people will just start using them.

There should be learning sessions, sharing knowledge about how prompt engineering works. This totally depends on use cases and domain knowledge. Sharing which kind of prompt engineering works in one scenario and doesn't work in another becomes very critical.

🔥 ChaiNet's Hot Take: The "start using AI" directive without proper training is setting teams up for failure. Cultural change and education must precede tool adoption.

Q: How should startup founders think about AI integration when building engineering teams?

Priyank: Definitely need to hire engineers based on their critical thinking and problem-solving skills. These skills are now more important than just knowing a specific coding language.

From day one, establish a culture of rigorous code review because all AI-generated code must be checked by humans to maintain high quality and security.

I see value in building simple solutions that actually address real-world problems rather than just generating complex solutions using AI. I remember one example – recently there was news that a developer wrote just 47 lines of Python code to summarize and classify customer tickets using AI, hosted it as a paid service, and made $2 million out of it.

🔥 ChaiNet's Hot Take: The million-dollar 47-line solution exemplifies the new paradigm: value creation through intelligent AI orchestration rather than complex code bases. Simple solutions that solve real problems win.

Q: Looking 2-3 years ahead, what capabilities should developers be building?

Priyank: Developers need to master AI-native architecture – learn how to design systems where AI is a core component, not just a tool. Develop deep product and domain expertise, understanding user and business goals.

We need to go beyond basic prompts and learn advanced AI interactions. This includes skills like model tuning, creating complex AI workflows.

What I see as critical is prioritizing learning about AI ethics and security because if you give AI a free hand, it can delete itself as well. We need to be very mindful of that.

🔥 ChaiNet's Hot Take: The mention of AI deleting production databases (the famous Replit case) highlights the critical importance of AI safety and governance in development workflows.

Lightning Round: Rapid-Fire AI Tips

Best AI tool for debugging complex issues: Gemini Code Assist – you can guide it through the problems.

One productivity hack using AI every developer should know: Recently I used Gemini CLI with a .gemini.md file in the root folder where you define coding styles and naming conventions. You don't need to worry about them anymore – they're passed as global context and save tons of time.

Biggest timesaver AI has given you personally: Researching and quick prototyping. I leverage Gemini's deep research capabilities.

One thing AI cannot do: Domain-specific critical thinking on tradeoffs to design the right architectures. Also, doing laundry – jokes apart!

Best resource for staying updated with AI: AI research blogs from leading companies. For summarized versions, Alpha Signal AI. I also find Medium articles resourceful.

Final Thoughts: The Future is Collaborative, Not Replacement

Priyank's journey from traditional coding at Microsoft to AI-powered development at Google shows us that the future isn't about AI replacing developers – it's about creating a new breed of superpowered engineers who can solve bigger problems much faster.

The numbers don't lie: 60-70% AI-generated code, 37% completion rates, and engineers focusing on strategy rather than syntax. But the human elements – critical thinking, domain expertise, ethical considerations, and architectural vision – remain irreplaceable.

As Priyank puts it: "Stay hungry, stay foolish." The tools are evolving rapidly, but the fundamentals of problem-solving, continuous learning, and hands-on practice remain constant.

The future of development isn't AI replacing humans – it's humans working with AI as the ultimate pair programming partner.

Connect with Priyank on LinkedIn for more insights into AI-powered development at Google. And remember, at ChaiNet, we learn from builders before the machines take over – or hopefully, before they become our best pair programmers.