AI Code Generation: Google's 75% AI-Written Code Claim

TL;DR

• Google's CEO, Sundar Pichai, says 75% of their code is AI-generated.
• This highlights AI coding tools' massive, practical adoption in big tech.
• It changes how we think about developer productivity and future software engineering.

The news in 60 seconds

So, Sundar Pichai, Google's CEO, just dropped a bombshell: 75% of the company's code is now AI-generated. That's a huge number, isn't it? He mentioned this recently, and it's not just about code completion anymore. This figure tells us Google's engineering teams are deeply integrating AI assistants and tools into their daily work. It matters because it sets a new bar for how much AI can actually contribute to a codebase, pushing beyond simple suggestions to generating substantial chunks of functionality. We're talking about a real shift in how software gets built at an enterprise scale.

Under the hood

When we talk about AI-generated code, it's not some magic black box. It's often about giving a clear prompt and letting the model, like Google's own Gemini, do the heavy lifting. Think about how many boilerplate functions or standard data transformations you write. An AI can whip those out in seconds. For instance, you might prompt for a simple data utility, and get something like this:

def calculate_average_price(items):
    """
    Calculates the average price of items in a list.
    Each item is expected to be a dictionary with a 'price' key.
    Returns 0 if the list is empty or no valid prices are found.
    """
    if not items:
        return 0
    total_price = 0
    item_count = 0
    for item in items:
        if 'price' in item and isinstance(item['price'], (int, float)):
            total_price += item['price']
            item_count += 1
    return total_price / item_count if item_count > 0 else 0

This isn't just auto-completion; it's generating a complete, documented function from a high-level request. And it's a pattern that scales, especially for common tasks, which Google probably has millions of.

Try it yourself

Want to get a taste of this? You can try a similar workflow with publicly available tools.

1. Pick an AI coding assistant: Grab a free trial for GitHub Copilot, Cursor, or even use Google's Gemini Pro API directly in an IDE extension.
2. Define a clear task: Think of a simple, self-contained function you need. Like,