Meet the AI That Only Makes Decisions (Jev)

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Meet the AI That Only Makes Decisions (Jev) · Channel: n8n · 04:45 · Watch on YouTube

Topics: AI agents, Task and workflow automation, AI models

Summary

The video introduces Jev, a model from Typesafe that returns probabilities for outcomes rather than natural-language responses, positioning it as a “smart if statement” for n8n. The presenter demonstrates routing based on a typed question, including filtering decisions by their confidence score. They then show how Jev can decide whether a deal needs executive review and triage incoming support cases, replacing workflows that might otherwise use an LLM followed by conditional logic. The video frames Jev as a faster, cheaper option for semi-deterministic workflows where pure logic is insufficient but AI need not handle everything.

Key points

Jev is presented as a new model category that outputs probabilities for outcomes based on input text and configured categories.
In n8n, the “route an item by typed question” option lets users define named outputs such as yes and no.
The example workflow routes a question to “yes” and gives a 97% probability for the outcome.
Confidence scores can be used to filter decisions, such as requiring at least 95% certainty.
A deal-review example shows Jev deciding whether a deal needs review from an executive.
The presenter says Jev is not a replacement for clear, deterministic rules.
The video compares using Jev in one step with using a large language model before an if statement.
A ticket-triage example demonstrates routing incoming Salesforce cases into categories such as critical, low, or high.
The presenter describes Jev as a fit for semi-deterministic workflows and says it is faster and cheaper than traditional LLMs.

Chapters

Start	Chapter	Summary
00:00	Introducing Jev	The presenter introduces Jev as a probability-based model that can work like a smart if statement in n8n.
00:33	Routing a typed question	A simple yes-or-no example shows how to define routes and use Jev’s probability score.
01:39	Deciding whether a deal needs review	The presenter configures Jev to evaluate deal data and explains how it compares with an LLM followed by an if node.
03:12	Triaging incoming tickets	A Salesforce case example demonstrates using Jev to route tickets into categories.
04:13	Where Jev fits and how to get started	The presenter describes Jev’s role in semi-deterministic workflows and points viewers to setup resources.

Quotes

It’s like a smart if statement. — 00:00

There is a 97 % chance that you are very, very cool. — 01:04

It’s all in one step and it’s vastly cheaper than anything we’ve seen like this before — 02:42

You just need AI to make a decision. — 04:13

more deterministic and much cheaper and faster than the traditional LLMs — 04:45

Concepts and entities

Jev, TypeSafe, n8n, Salesforce, workflow semideterminístico

Full transcript (original language)

00:00 not only did a new model just drop but a whole new model category and more than that it’s like it was purpose built perfectly to work with n8n i’m of course talking about jev the new model from typesafe it doesn’t respond with natural language like we’re used to it responds with probabilities on outcomes based on text we input and categories we set It’s like a smart if statement. Let me show you what that means in practice. As long as you’re on the latest n8n version, you can just search Jeff.

00:33 Not Jeff, Jeff. The memes are getting to me. As long as you’re on the latest n8n version, you can search “jev” and it’ll be right here. There’s a couple versions of it, but what kind of will make the most sense is route an item by typed question. And what that means is we can put in an input. So this is what it would actually take, right? So I’ll put in, I really like any then like a lot. Am I cool? The instructions will be answered the question, and then we can add in

01:04 routes just like we would in like an if statement in n8n. So route one can be yes and route two can be no. When we look at this on the canvas, we can see it has our named outputs there. So if we run this, It of course is going to say yes, because if you like any then, guess what? You’re cool. There is a 97 % chance that you are very, very cool. The reason why having the probabilities is so cool is let’s say that you have a really important use case and it can only pass through If you’re 95 %

01:39 sure you can then go down and filter again from this probability. Of course, this workflow is extremely silly. So let me show you what it looks like in something that looks a little bit more like an actual production workflow. So here we’re deciding if a deal needs a review from an actual person, from an executive. So we’re just going to plop. Well, actually copy this. the title of that, add “jev” right here. And in the instructions, we’re just going to say deal worth an exact review since

02:10 it’s already copied. I’ll paste it in here. And then the state is essentially what we’re passing in here. So the content to evaluate couldn’t have said it better myself. So we can say like name or whatever else, but we can actually also just pass in the full JSON. dot toJsonString. Since I’m in a hurry, I should be in bed already. And then that’s the state that it’s evaluating on. So then we can add a route. So we can just say, yes. Or… No. And now when we, well first we’ll delete the if node.

02:42 Thank you. Now when we actually run this, It goes in and chooses for you. Of course, you wouldn’t choose this if you already have deterministic logic there. That’s a clear yes or no or true or false. This is not better than that, of course, but this is so fast and cheap that it is way better than having a large language model before an if statement. to decide that and then having an if after. It’s all in one step and it’s vastly cheaper than anything we’ve seen like this before, especially considering the accuracy.

03:12 Let me show you what it looks like replacing a switch node instead. This is one of the biggest examples I’ve seen of this where it is triaging incoming tickets. So let’s recreate this. Essentially, it’s taking a new case in Salesforce and then this is putting in a bucket of if it’s critical, low, high, whatever. So we’re just going to add Jeff in yet again. I’ll actually copy this node name first. Add Jevin. And I’ll put that there. Instructions, and I’ll do the same thing again.

03:43 Thank you. where I do the little trick of replacing that with two JSON string. All that’s doing is all of this data that’s right here, is just plopping it right here for Jeff to see. That’s all that’s doing right there. You could also go like ID, and drag this and so on with each of them but i am not doing that because I’m lazy. So now we’ll add the route. This is what were these route names.

04:13 Okay. And now if I just. Yoink that right out of there and wire these up. I hope you’re starting to see exactly why this is getting so much attention, especially for n8n. This is so perfect. It fits that category of semi-deterministic workflows where You don’t want AI to do everything. You might just not be able to get a decision in perfect, pure logic. You just need AI to make a decision. And with this, there’s a full new category where it’s more

04:45 deterministic and much cheaper and faster than the traditional LLMs in a way that feels much more native. So I encourage you to go play around with it. The first few links in the description will be the resources you need to get started. So it will be how to get self-hosted with n8n, get a cloud trial. a Jeff API key because you’ll need an API key from Jeff to use this. If you have any questions, leave it in the comments. I’ll be down there because it is very late right now when I’m recording this, but I want you all to know about it. So I’ll see you in the next one.