Teaching a Cloud-Security Tool to Take Plain English
Searching a cloud infrastructure graph with Orca Security's Sonar used to be a language test: users had to construct JSON objects naming the exact model, navigate parent-child hierarchies, and pick from 18 operators such as equals or contains. That combination is what makes a domain-specific query language powerful and hostile at the same time — precise, and only usable by people who have already learned it. Orca partnered with AWS to build Discovery AI, a natural language query agent on Amazon Bedrock, so that the same search could start from an ordinary question. The companies report the result cut query costs by 90 percent, input tokens by 93 percent and latency by 75 percent compared with the initial baseline — the project's own starting point, which is a narrower claim than beating an established production system. The mechanism is the part worth remembering: the hard work isn't generating fluent text, it's reliably mapping that text onto someone else's schema.
AI-generated discussion · Different perspectives
The metrics headline is cost, tokens and latency, but the change that matters is who can ask questions. The whole premise is analysts who need fast answers without deep familiarity with a domain-specific query language - that's a staffing bottleneck being removed, not just a bill being trimmed.
Fair, though the headline numbers are still relative to the team's own initial baseline. A 90 percent cost reduction from a first working version and a 90 percent reduction from a system already in daily production are different bragging rights.