Same prompt through 4 AI planning tools: 8 vs 54 manual actions, a 6x gap
SSShken · reddit · 2026-09-12
The author ran the same one-sentence app idea (a single-page week planner) through four AI planning tools—OpenSpec, Spec Kit, BMAD, and Kiro—all executed by Claude Code with Opus 5, counting manual actions needed to get from "I have an idea" to "my agent can start." The gap is 6x: OpenSpec took 8 actions; Spec Kit took 4 to the first prompt and 14 to a full task list; BMAD took 10 and produced no task list at all; Kiro took 15 to the first prompt and 54 to a full plan—32 of which were clicking Allow on permission dialogs.
Other findings: three of four tools asked the same clarifying question about week format; none offered a way to edit tasks; all four added an unrequested delete function; the identical sentence yielded four different recommended tech stacks. Only Kiro required an account and admin rights, and Spec Kit's README ships a command that doesn't work until you find a release tag yourself. The author argues this action count is an overlooked metric and asks whether the gap holds on real, non-toy projects.
Related event: Same Prompt, Four AI Spec Tools: Planning Steps Differ by 6x(2 posts)→
More from coding & agent
- Researcher predicted multi-agent hidden coordination failure mode a year ago — it's now real — tianshi_li · 2026-09-12
- Dev uses AI to build a macOS widget for Fahrenheit-Celsius conversion, iterating past the 'AI slop' stage — floguo · 2026-09-12
- NoSpoon agent autonomously cranks out hilarious AI microdramas, 40-min episodes coming — Kyrannio · 2026-09-12
- Minecraft survival bot masters walking but keeps dying to night drifters; burrow goal next — zeeg · 2026-09-12
- LinkedIn user claims GPT-6 built a pixel-perfect Figma design system in 3 hours — AIandDesign · 2026-09-12
- OpenAI to co-host 'Agents, Everywhere' one-day hackathon across 50 cities on Sep 12 — seanmcdonaldxyz · 2026-09-12