Author(s): Leslee priyatham

Originally published on Towards AI.

A new decision model raised a more useful question: why are we asking text generators to make every choice?

You ask an AI model whether an email is spam.

After the opening example, the article argues that modern LLM tooling can be forced to output structured JSON, so the familiar “LLMs can’t return reliable structured data” critique is outdated. Instead, the real difference is purpose-built architecture: Jev (“System One”) is designed to produce bounded, typed decisions with confidence estimates rather than open-ended prose, which changes how developers should build real systems—especially around thresholds, logs, fallbacks, and inspecting mistakes. The author compares Jev’s workflow-oriented evaluation approach to what it can and cannot prove (reproduction of workflow judgments rather than universal correctness or policy alignment), critiques constrained-output “no hallucination” claims via the possibility of wrong choices from a fixed menu, and emphasizes that confidence numbers only help when validated on the application’s own data. Ultimately, the piece concludes that the useful question isn’t whether LLMs can be coerced into structure, but why software keeps routing every AI-shaped problem through the same text-generating model instead of selecting the component that matches the job.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI