From 25 % to 75 % in 18 months

Google announced that 75 % of new code created inside the company is generated by artificial intelligence and subsequently checked by human engineers. For comparison — in October 2024 it was approximately 25 % and in autumn 2025 Google reported 50 %. The pace of adoption is therefore exponential.

CEO Sundar Pichai emphasized in his post that Google is moving towards "truly agentic workflows". As a concrete example, he cited a complex code migration that a team of engineers and AI agents completed six times faster than would have been possible just a year ago with human labor alone. This is not a theoretical benefit — it is a measurable time saving in production code that runs services used by billions of people.

Engineers at the company primarily use their own Gemini models and some of them were even given specific targets in the area of AI utilization, which will be reflected in their performance evaluations this year. An interesting detail is that Google DeepMind employees were allowed to use the competing Anthropic Claude Code tool in recent months, which according to Business Insider created internal tension between teams.

The competition hasn't slept

Google is certainly not the only giant betting on AI in development. In April 2025, Microsoft CEO Satya Nadella stated that 20–30 % of code in some of the company's projects is written by artificial intelligence. His CTO Kevin Scott predicted at the time that within five years, 95 % of code will be generated by AI.

Meta goes even further in standardization. According to an internal document pointed out by Business Insider, the company has set a goal that 55 % of code changes in some organizations should be "agent-assisted". In the first half of 2026, it is expected that 65 % of engineers in Meta Creation Org will write more than 75 % of their code with AI assistance.

Similarly, Snap recently announced that according to its new operating model, at least 65 % of new code is generated by artificial intelligence. The trend is therefore clearly readable across all of Silicon Valley — AI is becoming a standard part of the developer toolchain, not an experimental gimmick.

What tools enable this and how much do they cost?

A wide range of code generation tools is available today for ordinary developers and companies. The most significant of them include:

GitHub Copilot — probably the most widespread tool, integrated directly into VS Code and other IDEs. An individual pays 10 USD per month (approximately 230 CZK), the team version costs 19 USD per user per month. Copilot supports dozens of languages including Czech in comments and prompts.

Google Gemini Code Assist — a direct competitor from Google that uses Gemini 2.5 Pro models. For individuals, it is available for free with limits, the full version for companies starts at 19 USD per month. The advantage is deep integration with Google Cloud and support for the Czech language in prompts.

Anthropic Claude Code — a tool that has gained popularity even inside Google DeepMind. It works as a terminal agent capable of modifying entire projects. It is part of the Claude Pro subscription for 18 USD per month (Max plan 100 USD per month).

Cursor and Windsurf — independent IDEs built around AI that are gaining increasing popularity. Cursor offers a free tier with limited tokens, the full version costs around 20 USD per month.

What does this mean for Czech developers and Europe?

For the Czech developer community, these numbers are relevant for several reasons. First, all the mentioned tools are available in the Czech Republic and Slovakia — no geographic restriction exists. Second, Czech is supported by modern models (Gemini, GPT-4o, Claude 3.7 Sonnet) so well that developers can write prompts and comments in their native language, even though production code remains in English.

The European context adds a layer of regulation. EU AI Act, which gradually came into force from 2024, emphasizes transparency and human oversight of automated systems. In the case of code generation, this means that companies in the EU will have to have clearly defined processes for checking AI-generated code, especially if this code affects critical infrastructure or personal data processing. Czech companies and startups should therefore already be building internal guidelines for using AI tools in development.

An interesting phenomenon is also the impact on the labor market. While some analysts warn of a decline in junior positions, the reality so far looks more like a transformation of the developer's role — from a pure coder to an architect and controller who manages AI agents and verifies their outputs. Just like Google, Czech companies are beginning to take AI adoption into account in requirements for new positions.

Where is this heading?

Google's announcement sent a clear signal: artificial intelligence in software development is no longer the future, but the present. When a company with almost 200,000 employees says that three quarters of its new code comes from AI, it is a benchmark for the entire industry.

An open question remains quality and sustainability. AI-generated code can speed up development, but it also brings risks — from security bugs to so-called "tech debt" to duplicate or inefficient implementations. That is why Google and the competition insist on human control. Technology is an extremely powerful tool, but the key remains in human hands. At least for now.