Why Write Code in 2026

22 days ago

Software engineers build and maintain the infrastructure (software factory) that enables agents to make changes and ship quickly.
Writing code remains valuable for direct thinking in the execution environment, enhancing attention, understanding, and ownership of system architecture.
Agents are not compilers but more like interns; they require precise context, and human involvement helps prevent fragility and slop in code.
Hands-on coding helps identify and fix weaknesses, establish architectural principles, and improve the software factory's robustness.
Details matter in software factories, from architectural patterns to algorithms; agents encourage evaluation and CI, but human intervention is essential for addressing weak spots.

The role of software engineers is to build the 'software factory', including the infrastructure that enables agents to successfully prompt for changes and ship immediately.
Proactive measures like prompts (skills, AGENTS.md, knowledge bases) and reactive protections through automated evaluation (tests, linting, type systems, evals) keep agents on track, allowing even simple models to produce good-enough changes.
While AI-generated code is prevalent, writing code remains valuable for direct thinking in the execution environment, fostering attention, understanding, and ownership, rather than passively observing agents.

Software architects must balance improving future code with maintaining current business speed.
Common architect failure modes include writing ignored documentation, enforcing rigid rules, or burning out by doing all the cleanup work.
Effective architects build systems that enable fast, safe contributions from all team members.

Good programmers hold the entire program in their head, like a mathematician holds a problem, enabling deep understanding and easy manipulation.
This ability is crucial early in a project to allow changing the problem being solved, not just the solution.
Factors that help: avoid distractions, work in long uninterrupted sessions, use powerful languages, rewrite programs often, write rereadable code, work in small groups, avoid multiple editors on the same code, and start small with prototypes.

The author still codes occasionally to stay familiar with the codebase, focusing on non-critical work like bugs and documentation.
Coding keeps the manager skilled and prevents skill atrophy, avoiding the disconnect of dismissing tasks as 'just a button'.
Leading by example through code sets standards for PRs, commit messages, tests, and code review communication.

The project watgo is a from-scratch Go project built with significant help from AI agents, contrasting with a previous Python rewrite.
Design began with a Markdown file, then agents wrote small, reviewable CLs (changelists); separate CLs for modifications and refactoring were used to keep changes manageable.
Projects are split into low-importance (vibe-coding) and high-importance (must review all agent code); watgo falls into the latter category requiring full human understanding.

The role of software engineers is to build the 'software factory', including the infrastructure that enables agents to successfully prompt for changes and ship immediately.
Proactive measures like prompts (skills, AGENTS.md, knowledge bases) and reactive protections through automated evaluation (tests, linting, type systems, evals) keep agents on track, allowing even simple models to produce good-enough changes.
While AI-generated code is prevalent, writing code remains valuable for direct thinking in the execution environment, fostering attention, understanding, and ownership, rather than passively observing agents.

Software architects must balance improving future code with maintaining current business speed.
Common architect failure modes include writing ignored documentation, enforcing rigid rules, or burning out by doing all the cleanup work.
Effective architects build systems that enable fast, safe contributions from all team members.

Good programmers hold the entire program in their head, like a mathematician holds a problem, enabling deep understanding and easy manipulation.
This ability is crucial early in a project to allow changing the problem being solved, not just the solution.
Factors that help: avoid distractions, work in long uninterrupted sessions, use powerful languages, rewrite programs often, write rereadable code, work in small groups, avoid multiple editors on the same code, and start small with prototypes.

The author still codes occasionally to stay familiar with the codebase, focusing on non-critical work like bugs and documentation.
Coding keeps the manager skilled and prevents skill atrophy, avoiding the disconnect of dismissing tasks as 'just a button'.
Leading by example through code sets standards for PRs, commit messages, tests, and code review communication.

The project watgo is a from-scratch Go project built with significant help from AI agents, contrasting with a previous Python rewrite.
Design began with a Markdown file, then agents wrote small, reviewable CLs (changelists); separate CLs for modifications and refactoring were used to keep changes manageable.
Projects are split into low-importance (vibe-coding) and high-importance (must review all agent code); watgo falls into the latter category requiring full human understanding.

Hasty Briefsbeta