- Author
- Birgitta Böckeler, Thoughtworks
- Published
- 15 October 2025
- Kind
- Practitioner article with hands-on trials of three tools, hosted on martinfowler.com
- Original
- martinfowler.com/articles/exploring-gen-ai/sdd-3-tools.html
- Subject
- Checkable specs for coding agents
- Themes
- Saying it exactly, Handing over control
What it is
Birgitta Böckeler is a Distinguished Engineer at Thoughtworks. The article is part of a series called Exploring Gen AI. It defines the term, describes three tools, and lists observations and questions.
What it says
- Böckeler separates three levels. They are spec-first, spec-anchored and spec-as-source.
- The article finds Kiro too heavy for a small bug. The bug became 4 user stories with 16 acceptance criteria.
- With spec-kit, Böckeler found many repetitive markdown files. The author would rather review code.
- The article reports the agent often ignored instructions. In one case it regenerated existing classes as duplicates.
- Böckeler says Tessl alone aims at spec-anchored work and explores spec-as-source. Generated code still varied between runs.
- The article compares spec-as-source to model-driven development, which never took off for business applications. It wonders if spec-as-source gets inflexibility and non-determinism.
What evidence it gives
The evidence is the author's own hands-on trials, mostly in September. Examples include one small bug in Kiro and one feature in spec-kit. The Tessl trial reverse engineered one JavaScript file. The article says it is not a close-to-real evaluation.
Interests and limits
Böckeler works at Thoughtworks. The trials were few and small. Spec-kit's feature was never finished. Tessl was in private beta, and the article says the tools change fast. The article says it lacks long-term usage reports.