Google CEO Sundar Pichai recently revealed that over 25% of Google’s new code is artificial intelligence (AI)-generated, a staggering statistic that underscores how large language models (LLMs) are reshaping software engineering. Let’s dive into how LLMs are reshaping the role of a software engineer.
Setting Expectations
LLMs excel at generating boilerplate code, summarizing and explaining concepts and assisting with unfamiliar languages and frameworks, though they often struggle with complex reasoning and deeply contextual tasks. Current generation LLMs are more like advanced autocomplete tools that select the most probable answer, not true intelligence. They sometimes hallucinate information and make mistakes, so human judgment remains essential. Their impact varies by experience level:
- Junior engineers can benefit by using LLMs to learn and explore unfamiliar technologies
- Mid-level engineers can use them to speed up mundane or repetitive tasks, and ask for code snippets or refactoring suggestions
- Experienced engineers may find limited value in tackling complex issues, but can leverage these tools for rapid prototyping.
Writing Code With AI Assistance
Software engineers integrate LLMs directly into their development workflows using specialized tools that enhance productivity and maintain code quality. Platforms such as Cursor and Continue.dev embed LLM capabilities within popular integrated development environments (IDEs), offering context-aware code suggestions, on-demand refactoring and rapid generation of boilerplate. These tools allow developers to save frequently used prompts, reducing repetitive input and speeding up interactions. They also enable seamless referencing of files in context, ensuring the AI understands the broader codebase and provides more accurate, relevant suggestions.
Additionally, features such as inline code generation, real-time error detection and automated documentation help engineers focus on higher-level problem-solving, while the AI handles routine tasks. By integrating these tools, developers can significantly reduce cognitive load and accelerate development cycles without compromising code quality.
LLMs Now Have Their Own Tools
LLMs are evolving from standalone code generators into integrated coding assistants equipped with dedicated tool suites. For example, Claude Code by Anthropic embeds directly into development environments, offering capabilities such as automated refactoring, debugging and inline documentation. Similarly, the model context protocol (MCP) has emerged as an open standard that enables LLMs to seamlessly connect with local repositories, databases and APIs, providing richer context for code generation and problem-solving.
OpenAI’s suite of tools, including APIs that support multi-step reasoning and interactive files or web searches, further enhances these models’ ability to contribute meaningfully to the development process. By leveraging these specialized tools, software engineers can access more accurate, context-aware suggestions and streamline tasks, reducing the cognitive load while accelerating development cycles without sacrificing code quality.
AI is Taking Over the Driver’s Seat
Emerging autonomous tools like Claude Code and Cursor Composer now empower developers to delegate entire, multi-step workflows — from initial code drafting and automated refactoring to debugging, testing and deployment — to agentic LLMs, allowing engineers to focus on higher-level strategy and complex problem-solving. Concurrently, the concept of ‘vibe coding’ (recently coined by Andrej Karpathy, former Director of AI at Tesla and OpenAI) is gaining momentum. This approach enables developers to leverage LLMs for rapidly generating exploratory code snippets that capture the essence of new languages, frameworks or libraries. It also accelerates the learning process and prototyping efforts, though the generated code is preliminary and requires thorough review to ensure quality and security.
LLMs are reshaping software engineering, evolving from advanced autocompletes to autonomous collaborators. While they boost productivity and spark innovation, human oversight remains key, especially in high-stakes software fields such as aerospace, healthcare and cybersecurity. As these tools continue to advance, the question is not just what we can build, but how we can build responsibly and securely.
