AI coding tools have undergone a dramatic transformation in just weeks, leaping from writing 20% of developer code to 80% in a single month. This shift represents a fundamental change in how software development works, moving AI from a helpful productivity tool to the primary driver of code creation. The acceleration is happening across the industry, with major tech companies already reporting that the majority of their new code is now AI-generated.

What Changed in AI Coding Capabilities?

OpenAI President Greg Brockman described the shift during a Sequoia Capital event, explaining how agentic coding tools, which use AI agents to autonomously write code, have moved from being a secondary resource to the main focus of development work. "We went from these agentic coding tools writing 20% of your code to writing 80% of your code," Brockman said, noting that this change occurred within December alone. "They go from being kind of a sideshow to being the main thing that you're doing," he told Sequoia partner Alfred Lin.

Greg Brockman

This isn't just happening at OpenAI. Alphabet CEO Sundar Pichai revealed that 75% of all new code at Google is now AI-generated and approved by engineers, up from 50% just last fall. At Anthropic, CEO Dario Amodei said engineers have already stopped writing code manually, predicting that AI will handle most software engineering tasks within six to twelve months. Former OpenAI researcher Andrej Karpathy went even further, stating he has not personally typed a line of code since December, delegating all programming tasks to AI agents.

How Are Companies Actually Using AI Code Writers?

The practical implementation of AI coding tools varies across organizations, but the pattern is clear: automation is replacing manual coding at an unprecedented pace. However, not all companies are adopting these tools blindly. OpenAI maintains a cautious approach, requiring human oversight before any AI-generated code is merged into production systems.

• Human Accountability: At OpenAI, a human must still sign off on all AI-generated code before it is merged, ensuring that someone remains accountable for the code in production.
• Speed vs. Understanding: Venture capitalist Chamath Palihapitiya warned that faster AI coding means little without capturing the reasoning behind engineering decisions, suggesting that velocity alone isn't the full picture.
• Continuous Improvement: OpenAI's latest model, GPT-5.5, is designed to be useful across foundational enterprise areas like agentic coding and knowledge work, showing that the company continues to refine these tools.

Brockman cautioned against blind adoption of AI coding tools, stressing the importance of maintaining human oversight. "We still want a human to be accountable for all code that gets merged," Brockman said. This approach reflects a broader concern in the industry about balancing speed with quality and responsibility.

Brockman

How to Implement AI Code Writing Responsibly

• Establish Code Review Processes: Implement mandatory human review of all AI-generated code before deployment, ensuring that at least one engineer understands and approves each change.
• Document Decision Rationale: Require AI tools to explain the reasoning behind code generation choices, not just produce working code, so engineers understand the logic and can catch potential issues.
• Monitor Quality Metrics: Track bug rates, performance issues, and security vulnerabilities in AI-generated code separately from human-written code to identify patterns and improve processes over time.
• Train Teams on AI Collaboration: Ensure developers understand how to work effectively with AI coding assistants, including how to prompt them effectively and review their output critically.

The shift toward AI-driven code generation is happening faster than many expected. Brockman is currently carrying additional responsibility at OpenAI, stepping in to oversee product after Chief of Applications Fidji Simo took medical leave, which underscores how central these developments are to the company's operations.

OpenAI's latest model release, GPT-5.5, represents another step forward in this evolution. The company released GPT-5.5 in April 2026 as its "smartest and most intuitive to use model" yet, with increased capabilities across multiple areas. Brockman claimed that the new model brings the company one step closer to creating a "super app," a multi-purpose unified service combining ChatGPT, Codex, and an AI browser into one integrated platform. The model is designed to be faster and more efficient than its predecessor, processing information with fewer computational tokens while maintaining superior performance.

The implications of this shift are profound. As AI tools become more capable at writing code, the role of software engineers is evolving from writing code to architecting systems, making design decisions, and ensuring quality. The companies that successfully navigate this transition, maintaining human oversight while leveraging AI's speed and productivity, will likely lead the next phase of software development.