AIToday
AI Coding AssistantsAI Safety & AlignmentZenn AI/MLPublished: Nov 22, 2026, 16:00 JST

Kiro's AI-added filter: who actually decided the spec?

Send on LINE
Kiro's AI-added filter: who actually decided the spec?

3 Key Points

  1. What happened

    Analyzing Kiro's 2025年7月 demo, the author notes Kiro turned a one-line request — add reviews — into a document that also included a review-filtering screen nobody asked for.

  2. Why it matters

    Treating AI output as a done deal means unrequested screens get designed, built and tested, so the useful check is whether a human reads the spec line and says 'not yet'.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

Kiro's craft-goods store demo serves as a starting point and works backward through requirements engineering history. Alistair Mavin and colleagues published EARS in 2009 for aircraft engine control systems, where the danger was that a dropped requirement still reads as a fine sentence — 'notify the user as needed' hides who decides 'needed', and an inspector cannot tell a fault from a situation that never arose. EARS keeps ordinary prose but constrains the order of conditions, events and responses, so undecided points become visible.

A second thread runs to OMG's 2001 MDA guide, which described generating implementations from models and admitted that early tools struggled to carry code-level fixes back into the model. Applied to reviews: if a deleted post is hidden by a quick code fix, a later AI rebuild from the old spec may bring it back. That is why the author stresses where a fix is recorded, not just whether code was generated.

The sharpest point is about intent. A test can pass under two readings of 'purchaser' — someone with an order history, or someone who received the goods — and only a concrete scenario about a cancelled order exposes the gap. The author hedges that the MDA lineage is not documented as a direct ancestor of today's tools; what the sources support is that machine-read designs and spec-code drift both predate generative AI.

FAQ
What is EARS, and why does it appear in an AI coding tool?
EARS is a 2009 requirements-writing technique from Alistair Mavin and colleagues, developed for aircraft engine control systems to stop builders and inspectors reading the same sentence differently. Kiro's announcement says it uses EARS, carrying a 16-year-old human-to-human technique into AI-assisted development.
Did AI invent spec-driven development?
No. Earlier precedents include OMG's 2001 MDA work on generating implementations from models, and BDD's scenario-based approach from the early 2000s. What is newer is that the tool itself proposes judgments that were not in the user's words.
How can you tell if a team is really doing spec-driven development?
The author cites Birgitta Böckeler's October 2025 observation that tools differ on how long a spec is kept: written per change, maintained continuously, or treated as the main human-edited source. The difference shows up at the second change request.
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleJerry Newman: AI's Big Winners May Not Be AI Companies