I’d Rather Be Unemployed Than Use AI for Coding – A 20‑Year Veteran’s Return to Hand‑Written Code

In 2022‑2023 the author, a software engineer named Brett with over 20 years of experience, began experimenting with AI‑assisted development tools such as GitHub Copilot, Cursor, and Claude Code. His first impression was that code completion distracted him rather than helped, because the generated snippets often appeared out of context and broke his thought flow.

A later encounter highlighted a classic AI hallucination: a colleague asked for help upgrading a dependency to version 4.6, and the AI confidently suggested that version existed. After searching, Brett discovered no such version, exposing the model’s tendency to fabricate plausible‑looking answers.

When Brett switched jobs in early 2025, management demanded that the team adopt AI for all coding tasks, citing a speaker’s claim that AI would transform knowledge work within six months. Brett reluctantly complied, trying Cursor’s Linux integration (which he soon abandoned) and then Zed’s built‑in AI. He experienced a dramatic shift: AI began to understand whole tasks, modify files, run commands, and iterate autonomously—what he calls “Agentic Coding.” Initially this boosted throughput, but over months he felt a growing existential anxiety, questioning the meaning of his work when an AI performed most of the implementation.

Beyond personal frustration, Brett observed broader risks. He noted that AI‑generated code often lacks the craftsmanship he values, making debugging and maintenance harder. He cited public data showing that each ChatGPT prompt consumes about 0.34 Wh (≈0.34 瓦时) and 0.322 mL of water, while Google Gemini’s requests use 0.26 Wh and 0.24 mL. Training large models such as GPT‑3 can require up to 70 kL of water per data‑center, and a single 20‑50‑turn conversation may indirectly consume 500 mL of water. A 2025 study estimates AI systems could use 3.125–7.646 × 10⁹ L of water annually, emitting 3.260–7.970 × 10⁶ t of CO₂, with downstream effects on hardware supply and consumer prices.

These hidden environmental and economic costs, combined with the danger of skill decay—engineers becoming dependent on AI and losing the ability to hand‑write, debug, and understand code—led Brett to a decisive conclusion. In spring 2026 he stopped using all AI tools, returned to writing code line‑by‑line, and even built a small game engine in his spare time, rediscovering the satisfaction of solving bugs and iterating manually.

The story sparked intense debate in the developer community. Some commenters warned that refusing AI could jeopardize employability, while others agreed that over‑reliance erodes the soul of programming. Notable voices included a veteran who coined the term “Code Ownershit” to describe loss of ownership, and references to Bjarne Stroustrup’s criticism of AI‑generated code quality. The discussion highlighted a tension between productivity gains and the preservation of craftsmanship, creativity, and ethical responsibility.

Ultimately, Brett’s experience serves as a cautionary case study for developers and managers: adopting AI tools without critical evaluation can lead to hidden costs, skill atrophy, and a diminished sense of purpose. He urges the community to prioritize quality, creativity, and human interaction over sheer speed, and to remain vigilant about the long‑term implications of AI in software development.