Live prompting interviews are hard: good prompts are subtle, and latency matters
brandon_galang · x · 2026-09-06
On live prompting as an interview loop: good prompts aren't flashy — they're subtly open-ended, delegate reasoning to the agent, and demand knowing when something looks off. The author used GPT 5.4 in Codex and spent much of the interview waiting on runs; he'd pick a lower-latency model like Gemini Flash 3.7 next time.
More from coding & agent
- Memory Trust Gap: agents follow stale memory 92-100% of the time, bigger models fool easier — rohanpaul_ai · 2026-09-06
- Memory Trust Gap: stale agent memory overrides fresh evidence, failures scale with model size — rohanpaul_ai · 2026-09-06
- Testing Astra: rebuilding Morrowind's atmosphere as a playable browser game in Babylon.js — kimmonismus · 2026-09-06
- Full system prompt revealed: how LLMs control a real robot arm via tool calls — YuXiang_IRVL · 2026-09-06
- Author rewrites AI coding guide with fresh Cursor, Devin and Conductor cloud coverage — vinvan · 2026-09-06
- Dev uses Astra agent to refine game combat while live-playing in a quick feedback loop — Dimillian · 2026-09-06