🔍 Read the full analysis: Could The Next Breakthrough In AI Be A Model That Doesn't Write Sentences? 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
TypeSafe AI has announced Jev, a novel AI model that outputs typed decisions rather than sentences. This development could reshape enterprise automation by replacing text generation with structured, decision-based outputs.
TypeSafe AI has unveiled Jev, an innovative AI model that does not generate sentences but instead produces structured, typed decisions with probabilities. This marks a significant departure from traditional large language models (LLMs) and could influence how enterprise AI automates decision-making processes.
Jev is described as a new class of AI, called System One Models, designed specifically for decision automation within software systems. Unlike standard LLMs that produce free-form text, Jev responds to structured questions with typed outputs, such as confidence scores and categorical decisions, streamlining integration into automated workflows.
The model is based on the concept of fast, intuitive decision-making, drawing from Daniel Kahneman’s System 1 thinking. It was developed by Diogo Almeida, a co-inventor of RLHF and InstructGPT, and received $40 million in funding led by DCVC. TypeSafe claims Jev operates at speeds of 70 to 500 milliseconds and at a cost of approximately $0.042 per million tokens, claiming it is significantly faster and cheaper than traditional models.
TypeSafe emphasizes that Jev aims to eliminate errors caused by output formatting and hallucinations, focusing instead on producing reliable, schema-compliant decisions. While it cannot invent off-schema answers, it offers a more deterministic and structured approach to decision automation, particularly suited to enterprise needs.
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
Implications of a Decision-Only AI Model
This development could fundamentally alter enterprise AI by shifting from text-based interactions to structured decision outputs. Companies could automate routine judgments with higher reliability and lower costs, reducing the need for human oversight in many decision-critical workflows.
By producing typed decisions with calibrated probabilities, Jev offers a more predictable and integrable form of AI, potentially reducing errors and increasing trust in automated systems. Its speed and cost advantages could expand automation to tasks previously deemed too complex or costly for AI.
However, this approach also raises questions about the limits of decision models, particularly regarding accuracy, scope, and the handling of nuanced judgments that traditionally require language understanding.
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 Toward Structured Decisions
Over the past three years, AI development has focused heavily on improving language models’ reasoning, context, and code generation capabilities. Major launches from OpenAI, Anthropic, and others have centered on enhancing conversational and generative abilities. However, these models still rely on free-form text, which can introduce errors and ambiguity in enterprise applications.
TypeSafe’s Jev represents a shift toward models optimized for decision-making rather than language production. The company argues that many internal business decisions—such as routing support tickets or evaluating risk—are better served by structured, calibrated responses rather than verbose text. This approach draws inspiration from cognitive psychology, specifically Kahneman’s System 1 thinking, emphasizing quick, intuitive judgments.
The concept of replacing reasoning-heavy, human-in-the-loop systems with fast, deterministic decision models is gaining traction, especially as organizations seek cost-effective, reliable automation solutions.
“Jev produces typed decisions with probabilities, not sentences. It’s more like a function than a conversation.”
— Diogo Almeida, co-inventor of RLHF
structured decision-making AI software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Limitations and Open Questions About Jev
While Jev’s decision outputs are designed to be schema-compliant, it remains unclear how well the model performs across diverse, real-world enterprise scenarios. Its accuracy, especially in complex or ambiguous cases, has shown variability in independent tests, with some results indicating overconfidence or underperformance.
It is also not yet confirmed how well Jev handles nuanced judgments that typically require language understanding or contextual reasoning. The model’s reliance on structured questions may limit its flexibility compared to traditional LLMs.
Further validation and real-world testing are needed to determine its robustness, scalability, and integration challenges in different industries.
decision-focused AI models for business
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Adoption and Validation
TypeSafe plans to release more detailed performance benchmarks and case studies in the coming months to demonstrate Jev’s capabilities in various enterprise settings. Industry adoption will depend on how well the model can handle complex decisions and integrate into existing workflows.
Further research will likely focus on improving accuracy, expanding decision types, and addressing limitations identified in initial tests. The company may also explore hybrid models combining decision outputs with traditional language understanding for more nuanced applications.
Regulatory and ethical considerations around decision automation will also influence how quickly and broadly Jev is adopted in sensitive domains.
As an affiliate, we earn on qualifying purchases.
Key Questions
How does Jev differ from traditional language models?
Jev produces structured, typed decisions with probabilities instead of free-form text, making it more suitable for automation and reducing errors related to output formatting and hallucinations.
What kind of decisions can Jev handle?
Jev is designed for three types of questions: choices, scores, and yes-or-no probabilities, which cover many routine enterprise judgments like support ticket routing or risk assessment.
Is Jev more reliable than existing models?
Jev aims to be more reliable by focusing on schema conformance and calibrated probabilities, but initial tests show variability in accuracy and overconfidence in some cases. Its effectiveness depends on proper question design and application context.
Will Jev replace all language models?
Likely not. Jev is specialized for decision automation within structured workflows, whereas traditional language models remain valuable for conversational, creative, and nuanced language tasks.
When will Jev be commercially available?
TypeSafe has announced the model and plans to expand testing and case studies in the coming months, with broader commercial deployment potentially later in 2026 depending on validation results.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
