Why write code in 2026
Direct code writing enhances attention, ownership, and architectural understanding in AI-driven software development.
Key Points
- Software engineers must maintain the software factory—building infrastructure that enables agents to operate effectively through proactive prompts and reactive safeguards like testing and linting.
- Even with AI-generated code, direct code writing provides deeper attention and understanding of system architecture, preventing fragility and slop accumulation.
- [AI Synthesis] Writing code fosters ownership and precision, allowing humans to experience the full 4DX virtual reality of execution, unlike passive reading of diffs or English descriptions.
- Humans benefit from writing code to think in executable, low-level or high-level languages, avoiding the limitations of English as an under-specified language for computation.
- Agents are not compilers—they act like interns, interpreting partial, possibly flawed code and generating changes; thus, human oversight is essential to avoid amplification of poor design decisions.
- When humans engage directly with code, they can identify and fix weak testing strategies, improve architectural consistency, and reduce long-term fragility.
- [AI Synthesis] The article argues that writing code helps humans think more precisely and avoid the ‘reverse centaur’ mindset—where decisions are made through English proxies that lack computational precision.
- A human who writes code can ‘joyfully delete’ and refactor, leading to better architectural outcomes than proxying decisions through English alone.
- The article highlights that agents often default to safe, conservative changes, which can amplify poor initial decisions—such as using local storage instead of a backend—leading to excessive code bloat.
- By directly engaging with code, humans can connect minute details to the big picture, improving system design and performance.