Issue #33: This AI doesn't write a single word. It just decides. And it's 200x faster than ChatGPT.

ChatGPT writes. Claude reasons. Gemini searches. Jev decides.

TypeSafe AI came out of stealth on September 15, 2026, with $40 million in seed funding and a model unlike anything in the mainstream AI conversation.

Jev doesn't generate text. It doesn't answer questions or draft emails.

It takes information you give it and returns a decision: a choice, a score, or a probability in milliseconds, at a fraction of the cost of a standard AI model.

It sounds narrow. The applications for small businesses are surprisingly broad.

Before explaining what Jev does, it helps to understand the problem it solves.

Every time you use ChatGPT or Claude for a task that involves sorting, classifying, or routing: "is this email a complaint or a sales inquiry?", "should this lead go to the high-priority or standard pipeline?", "does this invoice match our approval criteria?" you're using a Ferrari to pick up groceries.

Frontier language models are powerful, expensive, and slow relative to what the task actually needs.

They generate full paragraphs of reasoning to answer a question that only needs a one-word output.

Jev is built for exactly that gap.

TypeSafe calls it a "System One model" - a term borrowed from psychology (fast, intuitive thinking vs. slow, deliberate reasoning).

You give Jev a piece of information and a set of defined questions. It returns structured outputs: choices, scores, and probabilities.

No prose. No reasoning chain. Just the decision, with a confidence level, in under a second.

Classification — "Is this customer message a complaint, a refund request, a general enquiry, or a sales lead?" You define the categories. Jev reads the message and picks one, with a confidence score. 500 messages classified in about 40 seconds for roughly $0.02.

Scoring — "On a scale of 1 to 10, how urgent is this support ticket based on the customer's language?" Jev scores against your defined scale. Useful for prioritising queues, ranking leads, triaging content.

True/false and probability — "Does this invoice contain all required fields?" "Is this job application likely to be a good fit based on these criteria?" Jev returns a yes/no or a probability between 0 and 1.

Who built it: Diego Almeida, a co-creator of ChatGPT and key contributor to RLHF (the training technique that made ChatGPT useful).

He spent two years building a fundamentally different model architecture: non-autoregressive, meaning it doesn't predict the next word but scores candidates directly.

The result is 20–200x faster than a standard language model and $0.042 per million input tokens with free output.

For context on the price: 500 emails classified for lead type costs approximately $0.035. 700 support tickets scored by urgency costs approximately $0.009. 5,000 invoice checks per month costs approximately $0.21. These are not rounding errors. They're genuinely different economics from frontier model pricing.

The verdict: Jev is not a replacement for ChatGPT, Claude, or any generative AI model.

It's a complementary tool: specifically for the decision and classification steps in workflows where you currently either do things manually or pay frontier model rates for simple sorting tasks.

For small businesses with high-volume, repetitive decision points in their operations, Jev is worth knowing about.

For most small business owners without coding experience, it requires a developer or a no-code tool with Jev integration to use practically today.

That barrier is real and worth acknowledging.

A small e-commerce brand handling 200+ customer emails per day had a triage problem.

Their customer service team was reading every email to categorise it before routing complaints to one person, order status questions to another, refund requests to a third.

This categorisation step alone was taking roughly 2 hours of combined team time per day.

A developer friend spent half a day connecting their email inbox to Jev via the API.

Now Jev reads each incoming email, classifies it into one of six categories, assigns an urgency score from 1 to 5, and routes it to the right team member's queue automatically.

The team member opens their queue and sees pre-sorted, pre-scored emails. They start with the work, not the sorting.

Two hours of classification work per day: gone.

Cost of running Jev for their volume: approximately $3.50 per month.

Their developer friend billed them $300 for the integration, which paid back in less than a week of the team's recovered time.

If you have a developer or technical co-founder:

If you don't have a developer:

"If Jev is so much cheaper than ChatGPT, why wouldn't I just replace ChatGPT with Jev for everything?"

Because they do completely different jobs.

ChatGPT writes the email. Jev decides which folder it should go into.

ChatGPT drafts the proposal. Jev scores the incoming lead to decide whether the proposal is worth writing.

ChatGPT generates the customer service response. Jev reads the ticket and determines its urgency before anyone sees it.

They work in sequence, not competition.

The analogy: ChatGPT is a skilled writer at your company. Jev is the mailroom that sorts everything before it reaches the writer's desk.

You need both — just not for the same tasks.

The businesses getting the most out of Jev in early testing are using it to remove the classification and routing steps from workflows, so that generative AI and human judgment are only applied where they're genuinely needed.

The total cost picture for a small business: The API itself is nearly free at scale.

The real cost is the integration work, either a developer to connect it to your tools, or finding a no-code wrapper that already supports Jev.

For businesses with high-volume sorting problems (hundreds of emails, tickets, leads, or documents per day), the integration cost pays back quickly.

For businesses with lower volume or simpler workflows, the savings may not justify the setup effort right now.

This week: identify the one repetitive sorting task in your business that you or your team does manually at high volume.

Categorising emails. Scoring leads. Checking invoices for completeness. Rating support tickets by urgency.

Write down the decision criteria: what categories exist, what a high score looks like, what makes something true or false.

That documentation is your Jev question set, and it's useful regardless of whether you integrate Jev now or in six months when more no-code options exist.

Jev's launch is the clearest signal yet of a trend that's been building quietly: AI is splitting into two distinct categories.

Generative models: ChatGPT, Claude, Gemini, Fable 5 handle open-ended tasks requiring language, reasoning, and creativity.

Decision models like Jev and what will follow it, handle structured, high-volume classification at a fraction of the cost.

This split makes the economics of AI automation dramatically more accessible. A workflow that previously needed a frontier model for every step can now route simple decisions through Jev at $0.042/million tokens and reserve the $5/million model for steps that actually need it.

The total cost of running that workflow drops by an order of magnitude. For small businesses building any kind of automated process, understanding which steps need generation and which just need a decision is now a meaningful cost optimization question.

Using Jev for tasks that require nuance, context, or judgment that can't be pre-defined. Jev is excellent when you know exactly what the categories, scores, and criteria are in advance.

It struggles when the decision requires reading between the lines, understanding emotional context, or applying judgment that doesn't fit cleanly into predefined options.

A customer email that's half complaint, half compliment, and involves a complex order history, that's a ChatGPT or human call.

A straightforward "is this a complaint or a general enquiry?" across 500 emails: that's a Jev call. Know the difference before you build the workflow.

Until next issue, Kris

The Layman's AI — The only AI updates your business actually needs.