🔍 Read the full analysis: The Evolution Of AI: From Sentence Construction To Systemic Thinking on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
AI is evolving from text-based models to decision-oriented systems, exemplified by TypeSafe’s Jev. This shift aims to improve automation reliability and reduce reliance on traditional language models, impacting enterprise AI strategies.
On September 15, 2026, TypeSafe unveiled Jev, a groundbreaking AI model that departs from traditional language generation to focus on structured decision-making, signaling a major evolution in artificial intelligence for enterprise use. This development challenges the long-held assumption that large language models are the best tool for all AI tasks, especially within business automation, and introduces a new class of models optimized for decisions rather than text.
Jev is part of TypeSafe’s System One Models, designed specifically for decision automation inside software systems. Unlike conventional large language models (LLMs), Jev does not generate free-form text but provides structured, typed answers with associated probabilities, enabling software to act directly on these decisions. It handles three types of questions: choices, scores, and yes/no probabilities, returning results in milliseconds at a fraction of the cost of traditional LLMs.
The model is built on a new training approach called Reinforcement Learning for Calibrated Decisions (RLCD), developed by co-inventor Diogo Almeida, who previously contributed to RLHF and InstructGPT. TypeSafe claims Jev eliminates issues like hallucinations and overconfidence common in LLMs, as its outputs are schema-compliant and cannot invent off-schema answers, though it can still select incorrect options within the allowed set.
Economically, Jev is positioned as faster and cheaper than traditional models, answering in 70 to 500 milliseconds at roughly 4 cents per million tokens, making it suitable for high-volume, low-cost automation tasks. The company emphasizes that most enterprise decisions are System 1 judgments—quick, intuitive responses—making Jev a more fitting tool than reasoning-heavy models for many internal automation workflows.
Jev vs. LLMs: who should make the call?
Jev, from TypeSafe AI, is a “System One” model. It doesn’t write text. It returns a typed decision with a confidence score that your software can act on directly.
Same support ticket, two kinds of answer
“This ticket appears most likely related to billing, although it could also concern account settings or a recent plan change. I would suggest reviewing the invoice history before…”
A person reads it, or code has to parse the prose.
team: "billing"Software reads it and acts. Nothing to parse.
How they differ
| LLM | Jev | |
|---|---|---|
| Output | Text written for people | A choice, a score or a yes/no probability |
| Speed | Seconds per call | 70–500 ms* |
| Price | Input and (pricier) output tokens | $0.042 per million input tokens, output free* |
| Knows when it’s unsure | Often sounds confident when wrong | Confidence score on every answer |
| Explains its answer | Yes | No, which matters for audits |
| Best at | Reasoning, writing, open questions | Routing, tagging, scoring, duplicate checks |
* Vendor-reported. TypeSafe also claims up to 194× faster and 445× cheaper on its own selected workflows.
Accuracy is something you build
Jev is far cheaper and faster, but not more accurate than frontier models. How you phrase the question matters a lot.
TypeSafe’s benchmark scores agreement with two frontier models rather than verified ground truth. The five-question result used weights fitted on 1,000 labelled examples.
The real idea: a confidence dial you control
“duplicate listing”, confidence 0.62
Raise the threshold for fewer mistakes and more manual review. Lower it for more automation and more risk.
Only use Jev when all four hold
Good fits
- Routing tens of thousands of support tickets a day
- Flagging duplicate listings in a product catalogue
- Replacing a keyword filter that mis-tags half its matches
Poor fits
- Drafting customer emails or release notes
- Reviewing a few high-stakes contracts a month
- Anything that needs a written explanation
Impact of Decision-Oriented AI on Enterprise Automation
The shift from text-generating language models to structured decision systems like Jev could transform how companies automate routine tasks, reduce costs, and improve reliability. By providing calibrated probabilities and typed answers, Jev enables software to make faster, more consistent decisions without human intervention, potentially expanding automation into areas previously limited by the unpredictability of traditional LLMs.
This development challenges the dominance of large language models in enterprise AI, suggesting a future where decision-focused models complement or replace them for specific use cases. The move could also influence AI training methods, favoring calibration and schema compliance over free-form text generation, ultimately improving trust and accuracy in automated processes.
enterprise decision automation AI tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Evolution of AI Models and Enterprise Adoption
Over the past three years, major AI model launches have promised advances in reasoning, longer context windows, and improved code generation. Despite these developments, many enterprise applications still rely heavily on traditional LLMs, which often produce hallucinations, overconfidence, and unreliable outputs. The industry has recognized the need for more dependable AI systems, especially for decision-critical tasks.
TypeSafe’s Jev represents a strategic pivot, emphasizing decision accuracy and schema conformance. It builds on the understanding that many internal business decisions—like ticket routing, risk assessment, or compliance checks—are inherently System 1 judgments, suited better to rapid, structured responses than to free-form text. The model’s launch is part of a broader trend toward specialized AI models tailored for specific enterprise needs, moving away from the one-size-fits-all approach of large language models.
“Jev is designed to produce typed decisions with calibrated probabilities, making AI decisions more reliable for automation.”
— Diogo Almeida, co-inventor of RLHF and InstructGPT
Limitations and Performance Uncertainties of Jev
While Jev claims zero hallucinations by restricting outputs to schema-compliant answers, its accuracy depends heavily on how well questions are structured and the quality of training data. Benchmark results show variability, with some tests indicating lower accuracy than traditional models—62.6% agreement on phishing detection versus 81.3% for Claude Haiku 4.5. Additionally, its overconfidence in certain question types and underconfidence in others suggest calibration still needs refinement.
It remains unclear how Jev will perform across diverse real-world enterprise scenarios, especially those requiring complex judgment beyond the scope of predefined schemas. The extent to which Jev can replace reasoning models without sacrificing accuracy is still under evaluation.
Future Development and Adoption of Decision Models
TypeSafe plans to expand Jev’s capabilities and improve calibration through ongoing training and testing across various enterprise workflows. Wider adoption will depend on how well organizations can structure their questions and integrate Jev into existing automation pipelines. Industry analysts will monitor performance metrics and real-world case studies to assess whether decision-focused AI can scale effectively.
Further research and independent testing are expected to clarify Jev’s accuracy and reliability, potentially leading to new standards for enterprise AI decision-making tools. The company also aims to develop more specialized models tailored for different industry verticals, broadening the scope of automated decision systems.
Key Questions
How does Jev differ from traditional language models?
Jev produces structured, typed decisions with probabilities instead of free-form text, enabling direct software action rather than requiring parsing or interpretation of language outputs.
Can Jev completely replace reasoning-based AI models?
It is not yet clear if Jev can fully replace reasoning models. Its effectiveness depends on question complexity and context, with current benchmarks showing promising but not perfect results.
What are the main advantages of Jev for enterprise use?
Jev offers faster response times, lower costs, and more reliable, schema-compliant outputs that reduce errors caused by hallucinations or formatting issues in traditional LLMs.
What limitations does Jev currently face?
Jev’s accuracy varies depending on question structure and training data. It can still select incorrect options within its schema and requires careful question design for optimal performance.
What is the significance of this development for AI’s future?
This shift toward decision-focused models indicates a potential future where AI systems are more reliable, faster, and better suited for automation, especially in enterprise environments where decision accuracy is critical.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
