Vague prompts force AI to make endless calls: why good engineers still hand-write key code
bendee983 · x · 2026-09-18
Quoting John Crickett's take that good engineers will keep writing code by hand, this thread frames programming as a spectrum: from writing machine code (end 1) to giving vague high-level instructions (end 2). High-level languages pushed us away from end 1, and AI coding tools now bring us as close to end 2 as possible — you can vibe-code a full app from one sentence.
The catch: vague instructions force the AI to make many intermediate decisions that lack clear-cut answers. So real engineers using AI agents still specify granular details that matter — design patterns, architectural choices — even in natural language. As Petar Dambovaliev put it: the more precise you need to be, the more natural language it takes to describe, until at some point writing the code is simply faster than explaining it to an LLM.
Bottom line: AI coding agents are tools, not universal tools — knowing when to hand-write is itself an engineering skill.
More from coding & agent
- Dev grapples with Codex trust and security: all code is basically insecure — code_star · 2026-09-18
- Zhipu's ZCode caught silently uploading entire workspaces and full .git history to Aliyun OSS — terryyuezhuo · 2026-09-18
- ThursdAI Podcast Dives Into Jev Real-Time Models and Computer-Use Agents — altryne · 2026-09-18
- Real-Time Clippy Built With Jev Wakes Up Only When You're Stuck — yangyi · 2026-09-18
- ElevenLabs MCP adds voice, music, image and video generation in one workflow — Div_pradeep · 2026-09-18
- Reddit dev tests Opus 5.0 on system architecture: working code, poor design — True_Noor · 2026-09-18