In the rapidly evolving landscape of software development, senior developers are finding their roles transformed. As artificial intelligence continues to permeate the coding process, the responsibilities of these seasoned professionals are shifting from writing code to editing and evaluating it. This trend was highlighted by Addy Osmani, Google Cloud AI Director and former Chrome engineering lead, during an interview at the JS Nation US conference in New York. Osmani, a prominent figure with 14 to 15 technical books to his name, has consistently addressed the practical limitations of AI in coding. In this discussion, he sheds light on how senior engineers are becoming code editors rather than mere code writers.

The Current State of AI Coding and Declining Trust

According to Osmani, a staggering 90% of developers currently leverage AI in their coding practices. However, the reliability of these AI tools is on a downward trend. While AI proves effective for new projects or prototypes, it struggles with larger codebases and enterprise environments. Notably, the size of pull requests (PRs) has significantly increased, with AI frequently modifying unnecessary files or creating new implementations without reusing existing utility functions. Osmani's earlier concept of the '70% problem' remains relevant; AI can handle about 70% of the work, but the remaining 30%—which encompasses quality, consistency, and final touches—still requires human intervention.

Vibe Coding vs. AI-Assisted Engineering

Osmani distinguishes between two coding approaches: vibe coding and AI-assisted engineering. Vibe coding involves a free-form exploration of an idea's feasibility, allowing for rapid iteration. In contrast, AI-assisted engineering maintains traditional engineering principles such as architecture, security, performance, and quality while utilizing AI as a tool. The latter is essential for production code, where 'context engineering' plays a crucial role in determining the quality of the output. This technique involves providing the AI model with rich context, including documentation, examples, conversation history, and the structure of the codebase.

The New Role of Senior Engineers

Osmani emphasizes that the core role of developers is evolving from code writers to evaluators and editors. This shift is encapsulated in the provocative term 'highly-paid Code Editors.' Code reviews have become a critical area for training junior developers, where the ability to critically assess why AI chose a particular approach is more important than ever. Research indicates that engineers spend considerable time debugging AI-generated code that appears correct at first glance but is fundamentally flawed. Osmani describes this phenomenon as 'comprehension debt,' highlighting the cognitive load placed on developers.

Utilizing Background Agents

Osmani shares his practice of delegating tasks to agents via the GitHub app while on walks, allowing him to return to completed PRs. He recommends this method for small to medium-sized projects but advises against it in enterprise environments. He likens the evolution of this practice to transitioning from a 'conductor' managing a single agent to an 'orchestrator' overseeing multiple agents simultaneously.

Chrome DevTools MCP and Figma MCP

Set to launch by the end of 2025, the Chrome DevTools MCP (Model Context Protocol) will enhance coding agents by giving them 'eyes.' This advancement will enable agents to verify actual rendering results and utilize console logs and network information. When combined with Figma MCP, this will facilitate a workflow where design files can be implemented and validated against real screens. However, the technology has yet to reach a level where it can automatically recycle existing UI component libraries.

The Future of Browser AI and Trust Issues

Osmani points to the automation of user journeys as the next frontier, leveraging the rich context browsers hold—such as login information, calendars, and search histories. However, he stresses that maintaining human oversight is crucial in areas involving payments or personal data. He insists that automation should pause at stages that might raise user concerns, emphasizing the need for trust in design.

Advice for Junior Developers

Osmani advises junior developers to cultivate deep expertise in areas where AI has yet to excel, presenting a unique opportunity for differentiation. He counters the extreme view that programming languages or stacks are becoming irrelevant, asserting that understanding fundamentals remains a 'superpower' in the field.

Implications

The core message of this discussion is clear: in an era where AI writes code, the value of senior engineers lies not in their speed of writing but in their ability to read, judge, and provide context to code. The term 'highly-paid Code Editors' is not a diminutive label but rather a reflection of the essential skills demanded in this new landscape. As Osmani notes, while the capabilities of agents are improving, the costs associated with managing 'comprehension debt' remain unchanged. Bridging this gap will be a critical challenge for both tools and engineers moving forward.