Jev And Decision Modeling In AI: 24 Ways To Begin

🔍 Read the full analysis: Jev And Decision Modeling In AI: 24 Ways To Begin on ThorstenMeyerAI.com

Prime Big Deal Days · Oct 6–7Offer from Amazon

Get the latest gadgets delivered free — and shop member deals

• Fast, free delivery on millions of items
• Access to Prime Big Deal Days deals on October 6–7
• Prime Video, Amazon Music and more included

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

Thorsten Meyer published a guide on Sept. 29, 2026, mapping 24 potential uses for Jev, a tool for answering narrow, typed questions that software can use to make decisions. Meyer says three applications are live in his publishing operation, 12 meet his criteria for strong fit, seven need measurement first and two are poor fits.

Meyer describes Jev as a system that receives text or JSON alongside typed questions and returns answers in formats such as a yes-or-no probability, a choice among options, or a score on ordered levels. He says it does not write, summarize or extract information. Instead, its answers are intended to let a program branch on a decision without parsing a prose response. Meyer estimates a call takes 0.3 to 0.9 seconds and costs about $0.04 per million input tokens.

The central design principle is to automate clear cases and send uncertain ones elsewhere. In Meyer’s account, confidence is a signal for that routing: in a measurement involving 31 topics, he says Jev agreed with a frontier large language model 97% to 99% of the time at confidence of 0.8 or higher, and 42% of the time below 0.5. The source does not provide the test set, sample size or an independent evaluation, so these figures describe Meyer’s measurement rather than a general performance guarantee.

Meyer reports three live publishing uses: checking whether stories fit a site, detecting whether articles are in English, and classifying headlines when a primary model makes an error. He says a scan of 78,889 articles cost $2.01; it identified 1,576 non-English articles, of which 1,553 were fixed. For relevance, he reports roughly 10,000 story-and-site pairings judged over three days, with 22% clearly on topic. These are operational results he attributes to his own system.

A Test Before Automating Decisions

The guide outlines criteria for teams considering AI classification. Meyer says a use should have high volume, a narrow question, low-cost errors or a safe fallback, and an existing heuristic that has visibly failed. He advises keeping a keyword rule if it already works, unless evidence shows a need to replace it.

Meyer classifies same-event deduplication as a poor fit after a canary found zero duplicates to correct. He says a tool’s low cost and narrow task do not by themselves establish a need for deployment when the underlying problem has not been measured.

For higher-risk cases, Meyer’s examples include human review. A disclosure check, for instance, should send likely misses to a person rather than publish automatically. His proposed approach uses confidence thresholds and fallback rules: Jev supplies an answer, while the surrounding code determines what happens next.

From Publishing to Other Workflows

The article organizes its 24 examples across publishing, commerce, software, business operations and home use. The supplied source details the first six publishing examples, but ends partway through its commerce section. Meyer says publishing has the high volume and relatively low-cost errors he associates with Jev, while marking several proposed checks as unproven or unsuitable.

Among the publishing ideas, Meyer classifies disclosure detection and comment moderation as strong fits. For moderation, he proposes auto-approving clearly acceptable comments and hiding clearly identified spam only above a confidence threshold of 0.9, while queuing other cases. He marks a thin-source detector, product matching in roundups and headline-quality scoring as needing measurement first. Each example pairs a question with a rule, such as fetching an original source when an article lacks verifiable facts.

Meyer recommends replaying 300 to 500 past decisions, comparing results overall and by confidence band, and reviewing 20 disagreements to judge which answer was correct. He says teams should wire a use case in only where the high-confidence band reaches 95%, then use a separate feature flag, start with a 5% to 10% canary and expand from there. These are the author’s deployment recommendations, not reported results for every proposed application.

“Jev is the right tool wherever a system needs thousands of small judgements and can hand the unclear ones to something smarter.”

— Thorsten Meyer

Performance Claims Need More Detail

The source presents Meyer’s own measurements, but does not give enough methodology to assess how broadly they apply. It does not specify the sample size behind the 31-topic agreement result, how topics and confidence bands were selected, or whether the comparison with a frontier model was independently checked. The reported article scan and cost likewise lack details such as the model version, full input and processing costs, or how corrections were verified.

The article’s source text is incomplete: it stops during the commerce and customer-operations section. As a result, the remaining use cases, their individual fit labels and the basis for the total count of 24 cannot be reviewed from the supplied material. The seven cases marked for measurement also remain proposals until the relevant error rates and canary results are reported. It is not clear whether any of those uses have since moved into production.

Measure Before Wider Rollout

Meyer proposes that teams replay historical decisions, review disagreements, and use a small canary before expanding an application. For any use case, his stated milestone is evidence that the existing rule fails often enough to justify intervention and that Jev performs reliably on clear cases. The source does not announce a product release, a new deployment date or a planned follow-up report.

Key Questions

What is Jev?

Meyer describes Jev as a tool that takes text or JSON plus typed questions and returns structured answers, such as probabilities, choices or scores, for software to use.

How many Jev uses does Meyer say are already running?

Meyer says three applications are live in his publishing operation: site relevance checks, English-language checks and fallback topic classification.

What test does Meyer propose before using Jev?

He says a use should involve high volume and a narrow question, have cheap errors or a safe fallback, and address an existing heuristic that has been shown to fail.

Are Jev’s reported accuracy figures independently verified?

The supplied source reports Meyer’s measurement but provides no independent evaluation or enough testing details to establish how broadly the results apply.

Source: ThorstenMeyerAI.com