I Tested Three Spec-Driven AI Tools — source summary
A hands-on comparison (ranthebuilder.cloud, guest post by Itzhak Eretz Kdosha, Senior
SWE at Palo Alto Networks) of three spec-driven-development AI SDLC tools, run
against the same real feature (a backend addition to a serverless Python service).
Delivered via Telegram, ingested 2026-05-29. Text in raw/spec-driven-ai-tools.md.
Thesis
“AI coding tools make you faster. But speed without structure means you’re shipping ambiguity at scale.” Spec-driven workflows fix this: agree on what you’re building before the AI writes a line of code.
The tools compared
- OpenSpec (v1.2.0) — lightest footprint; you write delta specs (only what’s changing), which archive into a growing source-of-truth doc. Richest IDE integration (skills installed by default across 24 tools — see agent-skills). Frictionless iteration but no review gates; sometimes assumes context. Scored highest overall.
- Spec-Kit (v0.1.6) — GitHub’s spec-first tool. A project-wide “constitution” all
specs inherit; templates flag unknowns as
NEEDS CLARIFICATION. Strong planning artifacts, but implementation didn’t faithfully map to spec intent; no built-in code review; upgrade issues overwrite customization. - BMAD / BMAD Quick — best at course correction / iterative refinement and the only ones that added authorization permission checks; shared output directory by default (isolation configurable, not enforced). Healthiest project/community.
- (AWS AI-DLC noted but “too early-stage to evaluate.”)
Verdict & caveats
OpenSpec highest overall, but the ranking shifts with priorities — favor BMAD if you weight iteration/course-correction. Author’s caveat: these tools “land, break, and get patched between evaluation cycles,” so specifics date quickly.
Note
Agentic AI dev tooling; the unifying idea gets its own page, spec-driven-development, which connects to agent-skills.
Related
spec-driven-development · agent-skills