24 Ways To Work Through AI Decisions With Jev

🔍 Read the full analysis: 24 Ways To Work Through AI Decisions With Jev on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Thorsten Meyer published a guide on September 29 outlining 24 ways to use Jev, an AI decision tool, across publishing, commerce, software, business operations and the home. He says three applications are already running in his publishing operation, while 12 meet his criteria for a strong fit; the article presents these as his assessments and operating results, not an independent evaluation.

Meyer describes Jev as a system that receives a text or JSON state and a set of typed questions, then returns structured answers that software can use to make a decision. The answer types include a yes-or-no probability, a choice among options with probabilities and confidence, or a score on ordered levels. Meyer says a call typically takes about 0.3 to 0.9 seconds and costs about $0.04 per million input tokens.

In his account, three publishing uses are live: checking whether stories suit a site, checking article language, and supplying a fallback topic classifier when a primary model errs. Meyer reports scanning 78,889 articles for $2.01 in the language-check use, identifying 1,576 non-English articles and fixing 1,553. He also reports about 10,000 story-and-site judgments in three days for the relevance check, and 89% agreement with a frontier model for the classifier, rising to 97% to 99% for answers with confidence of at least 0.8.

The remaining examples are divided into 12 strong fits, seven that need measurement and two poor fits. Among the publishing examples, disclosure checks and comment moderation are labeled strong fits. Thin-source detection, product matching in roundups and headline quality need more measurement; deduplicating stories is a poor fit in Meyer’s canary because it found no duplicates. The supplied source excerpt cuts off during the commerce and customer-operations examples, so it does not provide the full list of 24.

A Test Before Automating Decisions

The guide’s central point is that a low-cost classification call is not automatically useful. Meyer argues a system should handle high volumes of narrow questions, where errors are inexpensive or uncertain cases can go to a more capable model or a person. He adds that teams should demonstrate a failing existing heuristic rather than assume one is inadequate.

That framework matters to teams considering automated checks because the suggested safeguard is to let software act only on clear cases. Meyer’s example routes low-confidence answers onward and keeps the prior process for the gray zone. He recommends replaying 300 to 500 past decisions, reviewing disagreements and enabling a feature flag on a small canary before wider use. Those steps are his proposed method; the article does not establish that they will produce the same results in other organizations.

How Jev Returns Decisions

Meyer says Jev does not generate prose, summaries or extracted passages. Instead, it returns typed responses to questions about supplied material, leaving the application code to determine what action follows. For example, a language check might ask whether an article is English, while a moderation check might select among options such as acceptable, spam, abusive or off-topic.

Confidence is a key part of Meyer’s case for using the tool. He reports a measurement on a 31-topic classification in which Jev agreed with a frontier model 97% to 99% of the time when confidence was at least 0.8, compared with 42% below 0.5. The source does not specify the sample size or provide an independent comparison, so those figures describe Meyer’s measurement rather than a general accuracy guarantee.

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

— Thorsten Meyer

Evidence Beyond Meyer’s Fleet

The reported costs, processing totals and agreement rates come from Meyer’s own publishing operation and tests. The source material does not give enough information to independently verify the measurements, including full sample details for the 31-topic comparison. It also does not establish how Jev performs across other industries, languages, data types or error costs.

The source excerpt ends partway through the commerce and customer-operations section. As a result, the remaining applications and their specific rules cannot be summarized from the supplied material. Meyer labels seven ideas “measure first,” but the excerpt does not include every measurement plan or a complete account of how the two poor-fit cases were selected.

Measure Before Wider Rollout

Meyer recommends testing proposed uses against 300 to 500 real past decisions, comparing results overall and by confidence level, and examining 20 disagreements to determine which system was right. He says teams should wire in a use only when the high-confidence band reaches 95%, then start with a feature flag off by default and a canary covering 5% to 10% of units.

The article does not announce a product launch or a timeline for expanding Jev’s use. Its next step for readers is to measure whether a specific workflow has a visible problem that fits the test, then trial the tool on clear cases while routing uncertainty for review.

Key Questions

What is Jev, according to Meyer?

Meyer describes Jev as a tool that returns structured answers to narrow questions about supplied text or JSON. Software can use those answers to route or handle a case.

How many of the 24 uses does Meyer say are already live?

He says three uses are running in his publishing operation: relevance checks, language checks and fallback topic classification.

What does Meyer’s fit test require?

His four conditions are high volume, a narrow question, low-cost errors or escalation for uncertain cases, and evidence that the current heuristic fails.

Are the reported accuracy figures independent benchmarks?

No independent benchmark is described in the supplied source. The 31-topic agreement figures are reported by Meyer as results from his own measurement.

What should a team do before deploying a Jev use?

Meyer recommends replaying hundreds of past decisions, reviewing disagreements, and testing with a small canary before expanding. He says to act only on high-confidence results and send uncertain cases for review.

Source: ThorstenMeyerAI.com