🔍 Read the full analysis: Jev For AI Decision-Making: 24 Ways To Put Models Into Practice on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Thorsten Meyer’s Sept. 29 article maps 24 ways to use Jev, a tool that returns typed, confidence-rated answers to narrow questions so software can make small decisions. Meyer says three uses are live in his publishing operation, 12 meet his fit test, seven need measurement and two are poor fits. The results and costs cited are the author’s own measurements, not an independent evaluation.
Meyer describes Jev as a system that receives text or JSON alongside typed questions and returns answers in forms such as a yes-or-no probability, a choice with probabilities, or a score on ordered levels. The output is meant to be consumed by code, rather than treated as prose to interpret. He says a call takes about 0.3 to 0.9 seconds and costs about $0.04 per million input tokens; these are figures reported in his article.
The three live uses Meyer identifies are a relevance check matching stories to sites, an English-language check, and a fallback classifier for 31 topics. He reports that a scan of 78,889 articles cost $2.01 and found 1,576 non-English articles, of which 1,553 were fixed. For the classifier, he reports 89% agreement with a frontier large language model overall, rising to 97%–99% for answers with confidence of at least 0.8. These results reflect his measurements and do not establish performance across other datasets or deployments.
The guide sorts the remaining proposals by fit. Meyer calls 12 strong fits, seven cases that need measurement first, and two poor fits. Among the publishing examples, disclosure checks and comment moderation are strong fits; headline quality and detection of thin sourcing need measurement. He labels same-event deduplication a poor fit after his canary found no duplicates to address.
24 use cases for Jev at a glance
Every use case, coloured by how well it fits
Proven in production
1Relevance gate: story and site2Language check3Classifier fallbackPublishing and content
4Thin-source detector5Same-event dedupe6Product fits the roundup7Disclosure present8Headline quality9Comment moderationCommerce and support
10Support-ticket routing11Return-reason coding12Review to feature complaints13Catalogue taxonomy14Order-fraud pre-triageSoftware and AI systems
15LLM guardrail16RAG passage filter17Citation check18Tool and intent routing19Log-line triage20PR risk triageBusiness ops and home
21Inbox triage22Expense categorisation23Lead qualification24Smart-home intent15 of 24 are ready to build or already running
A Test Before Automating Decisions
The guide’s central point is that low cost alone does not justify automation. Meyer says Jev is best suited to high-volume, narrow decisions where errors are inexpensive or uncertain cases can go to a more capable model or a person. A rule-based heuristic must also be shown to fail in practice; otherwise, adding a model may create complexity without fixing a measured problem.
For teams handling many repetitive checks, confidence scores could help separate routine cases from ambiguous ones. Meyer’s proposed pattern is to let software act on clear answers and keep existing processes for uncertain cases. That approach may limit the impact of mistakes, but its value depends on how well confidence corresponds to correctness in each real-world task.
The examples also give readers a way to judge the claims: the author distinguishes applications already in use from ideas that still need evidence. That distinction matters because the reported publishing results are an account from one operation, not independent validation or proof that all 24 applications will work elsewhere.
How Meyer Tests Jev’s Fit
Meyer’s four conditions are high volume, a narrow question that needs no multi-step reasoning, cheap errors or a route for uncertain cases, and a visibly failing heuristic. He advises measuring the existing process rather than assuming it has a problem. In his proposed trial, a team replays 300 to 500 past decisions, compares results overall and by confidence band, and reviews 20 disagreements to determine which answer was right.
He recommends wiring a use case into production only where the high-confidence band reaches 95%, then enabling it behind a separate flag and initially testing it on 5% to 10% of units. These thresholds and rollout steps are Meyer’s recommendations. His description emphasizes that the application’s code sets the action for each answer; Jev supplies a judgment, while the developer decides whether to accept, route or ignore it.
The article’s live examples focus on publishing, but its headline promises applications across commerce, software, business operations and the home. The supplied material details only the first part of the larger list, so it does not support a full account of all 24 proposals.
““Jev does not write, summarise or extract. You send it a state (text or JSON) and a set of typed questions, and it returns calibrated answers your code can branch on, with no prose to parse.””
— Thorsten Meyer, article author
Evidence Beyond One Publishing Operation
The available account does not include independent testing, detailed evaluation methods, or results from deployments outside Meyer’s publishing operation. It is also unclear how the reported agreement with a frontier model compares with human-checked accuracy, how the 31-topic set was assembled, and whether the cost and timing figures include all operating expenses.
Meyer says seven proposed uses need measurement because a failing heuristic has not been established, and two fail at least one fit condition. The supplied material names one poor fit, deduplication, but does not identify the second. It also ends before describing the rest of the commerce and customer operations examples and the other categories. Those details cannot be confirmed from the material provided.
Measure Before Wider Rollout
Meyer’s recommended next step for a prospective user is to replay several hundred real past decisions, inspect disagreements and verify performance within confidence bands. If the results meet his proposed threshold, he advises a limited canary rollout behind a flag. The article does not provide a scheduled follow-up, an independent review, or a timeline for testing the remaining use cases.
For the seven cases tagged “measure first,” the next milestone is evidence that the current heuristic fails often enough to warrant a change. Whether Jev performs reliably in other settings, and whether the remaining proposals will move into production, is not established in the supplied account.
Key Questions
What is Jev designed to do?
According to Meyer, Jev returns structured answers to typed questions about text or JSON. Software can use those answers to make decisions such as routing, filtering or prioritizing cases.
How many of the 24 uses are already live?
Meyer says three are live in his publishing operation. He classifies 12 other uses as strong fits, seven as needing measurement and two as poor fits.
What results does Meyer report from the article scan?
He says a scan of 78,889 articles cost $2.01, found 1,576 non-English articles and fixed 1,553. These are results he reports from his own operation.
Does the guide establish that Jev works for every listed use?
No. The article separates live deployments from proposed cases and says some need measurement. The reported results are not independent evaluations, and performance in other settings remains unclear.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
