AWS adds spec-check to Kiro

AI coding tools are good at writing what you asked for. The problem is that teams often did not ask clearly in the first place. That is the gap AWS is trying to close with a new Kiro feature called Requirements Analysis, announced May 12. Instead of checking code after an agent writes it, Kiro now tries to check the spec before the agent starts. ### What did AWS actually add? AWS added a Requirements Analysis step to Kiro, its spec-driven coding tool. Kiro already turns prompts into structured artifacts like `requirements.md`, `design.md`, and `tasks.md`. The new piece sits at the requirements layer and asks a stricter question: are these requirements internally consistent, and do they leave dangerous gaps? (geekwire.com) ### Why check specs instead of code? Because a lot of expensive bugs start before code exists. If a requirement is vague, contradictory, or incomplete, an AI agent will still produce something — but that “something” may include hidden assumptions nobody approved. Basically, the model fills in blanks with plausible-looking decisions, and those decisions can spread through the whole implementation. AWS is explicitly aiming at that failure mode. (geekwire.com) ### How does the new check work? Kiro uses a two-part setup. First, a large language model translates natural-language requirements into formal logic. Then an automated reasoning engine — an SMT solver — tests that logic to see whether the requirements conflict with each other or leave missing cases. That matters because the second step is not just pattern matching or vibe-checking. It is trying to prove whether the spec hangs together. (geekwire.com) ### What is an SMT solver in plain English? Think of it as a consistency engine for rules. If a spec says a user must be both over 18 and under 16 in the same path, the solver can flag that. If the spec defines one state transition but forgets another required case, the solver can flag that too. It is not proving the whole product is perfect — that would be a much bigger claim — but it can catch logical holes before they turn into code, tests, and production behavior. (geekwire.com) ### Why is this a Kiro thing specifically? Kiro has been built around spec-driven development from the start. AWS describes it as a tool that turns prompts into detailed specs, then code, docs, and tests. Its docs center the three-phase workflow — requirements, design, tasks — so adding formal checks at the requirements stage fits the product’s whole pitch. Other coding tools have added planning layers, but Kiro’s identity is more explicitly “write the spec first.” (geekwire.com) ### What else changed with Kiro? AWS also announced two speed features. Parallel Task Execution runs independent tasks concurrently instead of one by one, which AWS says can cut implementation time for large projects by about 75%. Quick Plan mode is the opposite kind of tradeoff — it lets developers skip the slower approval-by-step flow for features they already understand well, generating requirements, design, and tasks in one pass. (aws.amazon.com) ### Why now? The timing is not subtle. GeekWire tied the launch to broader scrutiny of agent reliability, including Amazon’s pushback earlier this year on a report linking AI coding tools to AWS outages. The announcement also landed just after Shawn Bice returned to Amazon to lead AI Services and the Automated Reasoning Group behind the feature. So this is product work, but it is also a signal: AWS wants to frame itself as the company putting harder guardrails around coding agents, not just making them faster. (geekwire.com) ### Bottom line? AWS is making a bet that the real bottleneck in AI coding is not typing code — it is nailing down intent. If that bet is right, the winning tools may look less like autocomplete and more like systems that force ambiguity to surface early, when fixing it is still cheap. (geekwire.com)