Where does Jev fit in your stack?

Before I get into how we’re using Jev, it’s probably worth explaining what it actually is.

Jev is a model from TypeSafe AI built to make decisions instead of generate text.

With GPT or Claude, you usually give the model a prompt and it writes something back.

With Jev, you give it something to look at and define the answers it can choose from.

For example, you could give it a lead and ask:

Is this person a fit for our ICP?

Then give it three options: yes, maybe, no.

Jev returns the answer along with how confident it is.

That’s basically what it does. It can classify something, score it, route it somewhere, or decide whether it should move to the next step.

We’ve started putting it in places where we have way more inputs than we realistically want a person looking through.

Take recruiting.

For one of our searches, we wanted candidates with at least two promotions across two different companies and roughly three years of average tenure.

If someone clears that, they answer five questions about how they use AI. Like:

How much of their work is already done with AI?

What tools are they using every day?

How much of their own money are they spending on AI each month?

Jev reads those answers with the role card and gives us pass, hold, or miss.

We only look at the passes.

If we’re hiring for a very small percentage of the people who apply, most profiles should never need a person to review them.

And Jev is pretty good at saying no.

We’ve started doing the same thing with content.

Our AEO workflow can generate around 40 ideas in a day. Some are things we’ve already covered. Some have an obvious AI tell in the opening. Some are worth looking at.

Jev scores them before they reach me.

Anything above 0.35 moves forward. Anything below it stays put.

I still open a few on either side of the cutoff every so often. If a 0.36 looks bad or a 0.31 looks good, then we know the threshold probably needs adjusting.

There’s also one rule we’ve been pretty strict about with this.

A pass does not mean send.

It does not mean publish.

It means move to the next step.

We might trust Jev to decide whether a candidate deserves a closer look. That doesn’t mean it gets to contact them.

Same with content. It can decide whether an idea is worth reviewing without getting permission to put it on the site.

That separation gets more important as you add more agents and more automated decisions.

I walk through seven of these builds here:

Watch the seven builds

If you build one, hit reply and tell me what you used it to filter!

Eric Siu