The release of advanced AI models like GPT 5.6 has sparked debate about the necessity of reading and writing code, but the speaker argues that human oversight in design, review, and quality control remains essential due to AI’s limitations in producing maintainable and reusable code. While AI can accelerate implementation, developers must continue to guide, trust, and manage code through established processes, balancing automation with responsible human involvement.
The recent release of GPT 5.6 and Anthropic’s Fable model has sparked renewed debate about the role of reading and writing code in software development. While some suggest that with advanced AI models, developers might no longer need to handwrite or even read code, the speaker argues that completely abandoning code reading is unrealistic. Most developers already do not read all the code they use, relying on trusted frameworks, libraries, and runtimes maintained by others. However, there remains a crucial distinction between not reading AI-generated code and not reading human-written code, with developers still responsible for selecting and trusting the tools they use.
The speaker presents a simplified hierarchy of software development roles, from junior developers who primarily implement code, to mid-level developers who design and implement, to senior engineers and tech leads who focus on design and delegation. Managers and directors typically do not write or read code regularly. This hierarchy highlights that design and architectural decisions are the core of software development, with implementation often delegated. The discussion emphasizes that while AI can generate large amounts of code quickly, it lacks the nuanced understanding to make maintainable and reusable designs without significant human guidance.
A key example illustrates the limitations of AI-generated code: an AI duplicated functionality across components instead of creating reusable ones, something no experienced human developer would do due to maintainability concerns. This highlights the non-deterministic nature of AI models, which can produce inconsistent results even under similar conditions. The speaker notes that while AI can assist with coding tasks, it requires careful steering and oversight by developers to ensure quality and maintainability, underscoring the ongoing need for human involvement in code review and design.
The speaker also addresses the cultural and organizational aspects of code ownership and responsibility. In large teams and companies, no single person can be responsible for every line of code, including CEOs who ultimately own the outcomes. Instead, systems and processes such as permissions, static typing, and robust frameworks help manage risk and maintain quality. The transition to AI-assisted development requires building these systems to minimize errors and maximize value, rather than relying solely on individual skill or micromanagement.
In conclusion, while AI models like GPT 5.6 and Fable are advancing rapidly and changing how developers work, the industry is still figuring out the balance between automation and human oversight. The speaker advocates for a middle ground where developers leverage AI to accelerate implementation but remain deeply involved in design, review, and quality control. This approach balances speed and reliability, acknowledging that fully relinquishing code reading or writing to AI is not feasible yet, but evolving workflows to incorporate AI effectively is the future direction.