Jev: an AI that doesn't write anything

Most AI tools you've heard of write things. You ask ChatGPT or Claude a question and you get paragraphs back. Jev, a new model from a company called TypeSafe AI, doesn't write at all. You give it something to look at and a fixed set of possible answers, and it picks one.

That sounds like less, and in one sense it is. For a surprising number of business jobs, though, picking one answer from a short list is exactly what you need.

What Jev does

You decide the answers in advance. "Is this email a complaint: yes or no?" "Which of these six categories does this enquiry belong to?" "Score this from 0 to 100." Jev gives you one of those answers and a measure of how confident it is.

The confidence figure is what makes it practical. TypeSafe's pitch is that you set a threshold: above it, the decision goes straight through; below it, a person takes a look. So the easy cases happen on their own and the awkward ones still reach a human.

It's in early access at the moment.

What we used it for

On Purpl, we used Jev to go back through the support tickets and categorise them, so we could see what the most common issue was. We ran it over more than 5,000 tickets. The whole job cost under $1.

Nobody was ever going to read and tag 5,000 tickets by hand. At under a dollar, you can simply ask the question.

Where the launch claims get carried away

TypeSafe's website leads with "193.6x Faster, 444.6x Cheaper" and "Zero Hallucinations". We'd treat both with care.

The page's own worked example puts Jev at $0.000081 and 0.114 seconds against standard AI models at $0.013880 and 8.566 seconds. That's roughly 170 times cheaper and 75 times faster: very impressive, and nowhere near the headline figures. The page doesn't say how "zero hallucinations" is measured either. A model that can only pick from your list can't invent an answer outside it, but it can still pick the wrong one.

None of that changes our experience: it was very cheap, and that's what makes it useful.

Jobs it suits

If a job is "look at this and put it in the right box", a decision model is worth trying. In a small business that might be:

• tagging incoming enquiries so they reach the right person
• sorting old support emails to find out what customers really struggle with
• flagging invoices or orders that look unusual so someone checks them
• scoring leads or applications against criteria you've written down

It won't write your replies, summarise a meeting or draft a proposal. For those you still want a model that writes.

Sorting jobs and writing jobs

Most talk about AI is about chatbots, but the useful work in a business is often a sorting job hiding inside a bigger task. Use a cheap, fast decision model for the sorting, and keep the clever, expensive writing models for the parts that need them.

If there's a pile of tickets, emails or forms in your business that nobody has time to sort, we build tools that do it.