Modern Engineering Values
17 hours ago
- #AI-assisted development
- #engineering workflow
- #coding agents
- The author's LLM workflow became outdated quickly, leading to a shift where coding is rarely done by hand, with AI writing most or all of the code for projects like Vite+, fate 1.0, Codiff, Athena Crisis, and Void.
- Coding agents now produce code as good or better than human-written code, significantly speeding up development and allowing work on multiple projects simultaneously, with a focus on reviewing and discussing code rather than writing it.
- The author uses Codex CLI with GPT 5.5 high for efficient solutions, emphasizing strong guardrails in the codebase, and employs spatial multitasking with dedicated screen areas for different projects to enhance productivity.
- Modern engineering values include strong ownership and domain expertise, where agents amplify ownership, and effective teams are small with clear boundaries; taste in deciding what to work on; strict guardrails and fast feedback loops to maintain velocity; and keeping context within the repository to aid agents and humans.
- Ownership of the stack is crucial as agents reduce dependency on third-party libraries, allowing full control over products; option value is emphasized to retain flexibility for future changes; and engineering management is expected to become more technical, requiring managers to retain domain expertise and provide technical leadership.
- The author's productivity has increased dramatically with agents, shipping more than twice as many commits and three times as much code compared to two years prior, highlighting that the bottleneck has shifted from writing code to exercising judgment, with a need for more people to leverage these advancements.