Jev And AI Decision Models: 24 Practical Ways To Get Started
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Jev And AI Decision Models: 24 Practical Ways To Get Started on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the latest gadgets delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

Thorsten Meyer published a guide on Sept. 29 mapping 24 potential uses for Jev, a tool that returns typed judgments for software workflows. He says three uses are live in his publishing operation, 12 meet his four-part fit test, seven need measurement, and two are poor fits; the source material provided details only for the first nine examples.

Thorsten Meyer published a guide on Sept. 29, 2026, mapping 24 uses for Jev, a tool that returns structured judgments to software systems. Meyer says three applications are live in his publishing operation, with about 90,000 decisions made so far; he classifies 12 other uses as strong fits, seven as requiring measurement and two as poor fits.

Meyer describes Jev as a system for sending a state, such as text or JSON, alongside typed questions. It returns answers in formats including a yes-or-no probability, a choice among options with probabilities and confidence, or a score on ordered levels. The intended use is to let software branch on the answer without parsing a prose response. Meyer says one call can carry multiple questions and typically takes about 0.3 to 0.9 seconds; he puts its cost at about $0.04 per million input tokens.

The three live examples are a relevance check for stories against a site profile, a language check, and a fallback topic classifier. Meyer reports that the relevance check assessed about 10,000 story-and-site pairings in three days, with 22% judged clearly on-topic. For the language check, he says a scan of 78,889 articles cost $2.01, found 1,576 non-English articles and fixed 1,553. His classifier fallback reportedly achieved 89% agreement with a frontier language model, rising to 97% to 99% when Jev’s confidence was at least 0.8.

The supplied guide also describes examples from publishing. It calls disclosure checks and comment moderation strong fits, while marking thin-source detection, product matching in roundups, and headline-quality checks as requiring measurement. A same-event duplicate detector is labeled a poor fit because, Meyer says, a canary test found no duplicates. The provided source cuts off during its commerce section, so it does not give details for the remaining examples in the 24-use list.

At a glance
reportWhen: Published Sept. 29, 2026
The developmentThorsten Meyer published a guide identifying 24 possible Jev applications and reporting that three are running in his publishing operation.

24 use cases for Jev at a glance

Publishing, commerce, software, business operations and the home, sorted by fit.

Every use case, coloured by how well it fits

Start in the green. Amber needs a measurement first. Red fails at least one of the four conditions.
livestrong fitmeasure firstpoor fit

Proven in production

1Relevance gate: story and site2Language check3Classifier fallback

Publishing and content

4Thin-source detector5Same-event dedupe6Product fits the roundup7Disclosure present8Headline quality9Comment moderation

Commerce and support

10Support-ticket routing11Return-reason coding12Review to feature complaints13Catalogue taxonomy14Order-fraud pre-triage

Software and AI systems

15LLM guardrail16RAG passage filter17Citation check18Tool and intent routing19Log-line triage20PR risk triage

Business ops and home

21Inbox triage22Expense categorisation23Lead qualification24Smart-home intent

15 of 24 are ready to build or already running

3
12
7
2
Live
Strong fit
Measure first
Poor fit
Live: in my fleet today. Strong fit: meets high volume, narrow question, cheap errors and a visibly failing heuristic. Measure first: the failing heuristic is unproven.
From “24 Ways to Use Jev” on thorstenmeyerai.com. Figures are my own production measurements, September 2026, rounded, unless marked illustrative.

A Test Before Automation

The guide sets out four conditions for considering a low-cost judgment: high volume, a narrow question, manageable error costs and evidence that an existing heuristic fails. Meyer advises retaining a simple rule when it works.

For these workflows, Meyer recommends acting on clear answers and routing uncertain cases elsewhere. He proposes replaying 300 to 500 past decisions, comparing results overall and by confidence band, then reviewing 20 disagreements. He says teams should connect a use only when the high-confidence band reaches 95%, start with a separate flag disabled by default, and canary it on 5% to 10% of units. These are recommendations in the guide, not independently verified performance standards.

Products Worth Considering

Kindle Paperwhite Signature Edition 32GB (2024 Model) - Metallic Black

Kindle Paperwhite Signature Edition 32GB (2024 Model) – Metallic Black

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite

As an affiliate, we earn on qualifying purchases.

Jev’s Role in the Workflow

Meyer describes Jev as a tool that provides typed answers for code to use, rather than writing, summarizing or extracting information. A yes-or-no probability may support a gate, a choice may route an item to a category, and a score may rank fit, quality, severity or priority. The application’s code determines what action follows.

Meyer reports a measurement on a 31-topic classification in which Jev agreed with a frontier language model 97% to 99% of the time at confidence of 0.8 or higher, compared with 42% below 0.5. The source does not specify the sample size or further test conditions for that comparison. Meyer recommends preserving the existing path for uncertain cases rather than assigning every answer an automated action.

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

— Thorsten Meyer, in the Sept. 29 guide

Evidence Still Needed

The supplied source does not include the complete 24-item list: it ends as the commerce and customer-operations section begins. Details for the remaining examples, including their questions, fit labels and decision rules, are therefore unavailable here. Meyer’s reported results are claims from his own guide; the provided material does not include independent evaluations or full test methods.

Several listed applications are explicitly provisional. The guide says the thin-source detector needs evidence that its replacement improves on a 300-character rule, and the product-fit and headline checks also need measured error rates. The duplicate detector remains a poor fit in Meyer’s account because its canary found no duplicates. The source does not say how often those tests will be repeated or whether the classifications will change.

Measure Before Wider Use

Meyer recommends replaying historical decisions, inspecting disagreements and comparing accuracy across confidence bands for proposed uses. Teams would then enable a separately controlled canary on a small share of units and expand if the measured results support it. The guide recommends routing uncertain cases for review and automating only answers that meet the selected threshold.

The provided material does not identify a follow-up release date, an external evaluation or a schedule for publishing results from the other use cases. It describes measurement by teams as the next step before connecting a proposed application to a live workflow.

Key Questions

What is Jev designed to do?

It returns structured answers, such as a yes-or-no probability, a category choice or a score, so software can use them in a workflow. Meyer says it does not write or summarize content.

How many applications does Meyer say are ready or running?

Meyer says three are live and 12 more meet his four-part test, for 15 live or strong-fit uses. He labels seven as needing measurement and two as poor fits.

What does the four-part fit test require?

The proposed use should involve high volume, a narrow question, low-cost errors or a route for uncertain cases, and a heuristic that has been shown to fail.

What evidence does the supplied guide provide?

It reports results for three live publishing uses and gives examples from the next section, but the supplied text ends during the commerce discussion. The performance figures are Meyer’s reported results, and full test methods are not included.

Source: ThorstenMeyerAI.com

You May Also Like

The Enforcement Countdown: 89 Days Until the EU AI Act’s GPAI Penalty Phase Begins

On August 2, 2026, the EU enforces penalties on GPAI providers under the AI Act, marking a key compliance deadline for AI companies operating in Europe.

Innovating Watercolour Art With AI: The Role Of TRL And OpenEnv

An independent engineer has fully recreated Surya Narreddi’s watercolour AI model using TRL and OpenEnv, releasing all artifacts openly on Hugging Face.

How To Make Sense Of Deep Learning In The AI Tools You Use

A guide to how deep learning trains AI tools, what the process explains about their strengths and limits, and why model size alone is not enough.

12 Innovative AI Devices Redefining Home Automation In 2026

Discover 12 innovative AI-powered home automation devices redefining smart living in 2026, emphasizing privacy, customization, and advanced control.