Jev will 10x your Claude Code (Here's How)

0:00There's a new AI model in town from the

0:02co-inventor of Chat GPT and it is

0:04insanely cheap and incredibly fast. It's

0:06called Jev. And just to show you how

0:08fast it is, I'll send this prompt and

0:11that is real time. It gave me an output

0:13for less than a second and at a fraction

0:16of the cost. So today I'll explain Jev

0:18for you simply and also some of the best

0:20ways by which you can integrate Jev with

0:22the agentic harnesses you use like cloud

0:24code to make your setup faster, your

0:26systems cheaper, and even build new

0:28things and automate parts of your

0:29business that weren't possible before.

0:32Let's dive into it. So first of all,

0:33what is Jev? And I won't go super deep

0:36into this, but basically Jev is a new AI

0:38model that is quite interesting because

0:40it was released by a co-inventor of

0:42ChachiBT. So this person Dooo, he made

0:45this post that already has something

0:47like 38 million views. And he says here

0:49that Jev is a new type of frontier AI

0:51model that is 20 to 200 times faster and

0:5540 to 400 times cheaper. And if you look

0:57at the rate card for this model, that is

0:59indeed the case. It is around 24 times

1:01cheaper than Haiku and around 230 times

1:04cheaper than Fable 5.1. And a big part

1:06of why that is is because its output

1:08tokens, basically its response to you as

1:11the user, is always free. and they only

1:13charge for the input tokens, which is

1:15essentially your prompt. And it's super

1:17cheap. It's only four cents per million

1:19tokens. Now, a big part of why it's so

1:21fast and so inexpensive is because Jev

1:24can only answer in three shapes. So,

1:26when you ask it a question, it can

1:27either give you a binary response

1:29whether that statement is true or false.

1:31It can provide a selection against a

1:33menu of options, or it can also give you

1:35a response that is based on scale. Let's

1:38say from 0 to 10. And even though this

1:39sounds like a big limitation for the

1:41model, this is actually the genius

1:43behind why it's so effective in the use

1:45cases that we'll go through later. But

1:46before we go to that, one key thing to

1:48remember is that Jev is not actually a

1:51large language model. And the way that

1:53Typesafe, which is the company behind

1:54Jev, talks about this is that they're

1:56saying that Jev is the first system one

1:58model, whereas all the other AI models

2:00like Fable, Astra, and the others are

2:03what they're calling system 2 models.

2:05Now, in case you've read this book

2:06called Thinking Fast and Slow, that

2:08system one and system two dichotomy of

2:10how humans and people think might be

2:12familiar to you, but essentially the

2:14difference between these systems is that

2:16system one is all about thinking fast

2:18and making snap decisions. And so, Jev

2:20as an AI model just optimizes against

2:22this and so it just outputs

2:24classifications and can do that really,

2:26really fast and really, really cheaply.

2:28Whereas LLMs and other general purpose

2:30models like Fable or Astra, they can

2:32output text and they do that by writing

2:34each word one by one, but it gives them

2:36more flexibility of what it can output

2:38obviously. But that would have the

2:40drawback of these system two models

2:42thinking slower versus its system one

2:44counterparts. And so really if there's

2:46one takeaway from all of this, I think

2:47the best way that you can use Jev right

2:49now is to combine both. Combine a system

2:51one model like Jev with a system 2 model

2:54like the claude models. And so I'll show

2:56you some use cases of how you can get

2:58started and actually get value from this

2:59today. But first, let's get you set up

3:01so that you can actually use Jev. And as

3:03with any other AI model, it's actually

3:05available through a variety of

3:06platforms. You can obviously use it by

3:08connecting to Typesafe, who is the

3:10company behind Jev. And as per their

3:12expost at the time of this recording

3:14today, they just announced that Jev is

3:16now actually available to everyone

3:17because previous to this just hours ago,

3:19there used to be a wait list to access

3:21it. So, if you go to this URL, you'll be

3:23able to sign up there and actually get

3:25your API key to connect it to Claude. At

3:27least when I was testing it personally

3:28and throughout the use cases that I'll

3:30go through here, I connected to Jev via

3:32open router, which is the service that

3:34always gets updated with the newest AI

3:36models as they get released. So, you can

3:37just access that through this URL. And

3:39so, to set it up with cloud or any

3:40agentic harness that you're using, it's

3:42just one prompt away as usual. And you

3:44can just take a screenshot of this if

3:45you need a starter prompt to set that

3:47up. Or if you want the prompts and the

3:48setup guide for everything that I'll

3:50cover here in this lesson. I also made

3:52this PDF guide which you can just grab

3:54for free below and you can just send

3:55that to your agent for all of the good

3:57nuggets that you can pick up in this

3:59video. So once you set that up and you

4:00confirm with Claude that you have access

4:02to Jev like what I did here, now we can

4:04get to actually using it. And I'll

4:06actually talk about this through three

4:07levels by which you can use Jev. And by

4:09the way, if you want to learn how to

4:10build and sell AI systems that

4:12businesses actually pay for, then that's

4:14pretty much all we do over at the

4:16Robbernuggets community, where not only

4:17do you get access to the Claude Living

4:19Master Class, which we update every week

4:21and takes you from zero to mastery with

4:23the latest on AI, but you also get

4:24access to our agents as a service

4:26course, which walks you through how to

4:28actually get paid for all these AI

4:30skills that you are learning. You also

4:31get to be part of a genuinely great

4:33community of AI builders. In fact, you

4:35can see just some of the recent wins our

4:37members are getting from the program

4:38right here. So if you want to start

4:39earning from AI then check that just in

4:41the pin comment below. Now back to the

4:42video and the first one is to integrate

4:44Jev with your own agentic operating

4:46system. Basically the way you work with

4:48your agents so that you can get faster

4:50results and cheaper systems. So less

4:52token burn. Now because the way we use

4:54agents differ depending on the work that

4:56we do. I'm sure that you can also find

4:57ways to use Jev outside of what I'll

4:59talk about. But just to give you an idea

5:01here are two use cases that I am testing

5:03out so far using this model. The first

5:05one is around model routing which is

5:07basically letting Jev automate the

5:09choice of the model depending on the

5:11task that we are giving Claude. And this

5:13is important because remember it is not

5:15really practical for you to use Fable

5:17all the time because out of all the

5:18models that is the most expensive. Same

5:20thing with Opus. If you just default to

5:22Opus every time then that can also drain

5:24your usage quite a lot. And for a lot of

5:26tasks sometimes Sonnet and Haiku which

5:28are the cheaper models are actually

5:30enough. But the problem there is for you

5:33to switch through these models and

5:34decide the right model for each task.

5:36That decision usually lies with you as

5:38the user. And so there wasn't really a

5:40quick and cost- effective way for us to

5:42automate model routing up until Jev. And

5:45so to set this up, you can just use this

5:46prompt for you to get started. And just

5:48to give you a visual demo of the test

5:50that I set up, essentially what I asked

5:52Cloud to do is to do a comparison of

5:54around 12 prompts with Jev and another

5:56one where it's running with Fable 5.1

5:58every time. And you can see here that

6:00because of Jeb and the fact that it's

6:02actually routing to the right model

6:03depending on the task, it actually

6:05resulted to 70% savings because nine out

6:08of those 12 tasks never needed the top

6:10auto anyway. So that is quite useful.

6:12But obviously you have to try it out for

6:13the work that you do specifically just

6:15to see if the output that you are

6:16getting is still good enough in exchange

6:18for the tokens that you are saving. But

6:20it's just great that we now have this

6:22new class of AI models that can actually

6:24do these types of decisions for us. Now,

6:26in practice, if you're testing this out,

6:28I do advise you to make a skill command

6:30first where you can switch jev off or

6:32on. For example, here in this Claude

6:33session, you can see I typed in /jv on.

6:36And so, for this whole session, whenever

6:37I assign it tasks, Claude will now use

6:39Jev in order to find the right model for

6:41that task. One example of that is this

6:44where I ask it to find the file path

6:45where the Jev router script lives. You

6:47can see that for that task, Jev actually

6:49assigned a Haiku helper, which is the

6:51cheapest model to find at file path,

6:53which is much more efficient for your

6:55token usage because if Jev wasn't there

6:57routing to Haiku, then we would have

6:59used Opus 5 here, which is the default

7:01that I'm using for this session. The

7:03second use case is making Claude become

7:04more efficient when finding the right

7:06skills. So again, this is just a visual

7:08demo of a test that I ran. But

7:10essentially what this shows is 14 tests

7:13where if you send it a prompt and you

7:15ask it to find a specific skill in my

7:17workspace, you can see here that Jev

7:19takes much less time to find the right

7:21skills versus if you just default to

7:24something like Opus 5, for example. And

7:26so in total for those 14 tests, Jev was

7:29able to find the right skill within 5

7:30seconds while Opus 5 took around 30

7:33seconds. And just to show you how Jev

7:35was used in this specific use case,

7:37basically your input is the task that

7:39you are trying to do. Jev then looks at

7:40that and the options that it can choose

7:42from would be your skills itself. So at

7:45least for me, if you can see, I have

7:46something like 145 skills in my

7:49workspace. And so because Jev is really

7:51quick, it can almost instantly output

7:53the right skill from that list which

7:55Claude then loads. And so if you want to

7:57test that out for yourself, then you can

7:58just copy this prompt and send it to

8:00your agent. Now, beyond just level one

8:02of giving you faster and cheaper

8:03systems, if we get to level two, this is

8:06actually how we use Jev for more

8:08business use cases because with Jev, you

8:10can actually make automations that are

8:11almost at lightning speed and doesn't

8:13cost as much as the other AI models. And

8:16just to give a visual demo, let's say

8:17you have a 100 emails and the automation

8:19that you are building needs to answer a

8:21business question, which for this case,

8:23we want to know which of these emails

8:24are actually leads that we can contact.

8:26But obviously it can be others like if

8:28these are customer support tickets then

8:30you can triage which ones are needing

8:31the most support. But at least for this

8:33demo what we'll show is Jev doing the

8:35classification here in this column. And

8:37then we'll also use Haiku as well as

8:39Fable in order to show the difference

8:41between the speed and cost of these

8:43models. So when I click run, these will

8:45now show the time it took for these

8:47models to classify each of these emails

8:49in full. So let's go ahead and run that.

8:52And as you can see that took Jeff like

8:53no time at all. within less than a

8:56second, it was able to classify all of

8:58those leads, whether they're warm,

8:59whether it's not a lead, whether it's

9:01cold, which is much faster and much

9:03cheaper versus these other models. And

9:05so, when it comes to business

9:06automations, that is where Jev really

9:08shines. If you have a huge volume of

9:10things and there's a business question

9:12that's associated to those things, then

9:14this is a good candidate for you to use

9:16Jev in. So, for example, in enterprise,

9:18there's a huge industry with regard to

9:20detecting invoice fraud. Spam detection

9:22software also has a good use case for

9:24this. Community moderation is another.

9:26Same when it comes to high volume

9:28requests for any refunds and even

9:30classifying your customers if they are

9:32churning or not if you're running a

9:33subscription software business for

9:35example. And so that's the pattern that

9:36I think would be good for you to think

9:38about in your company or in your

9:39business. What are the things that you

9:41are receiving in volume that you need to

9:43classify? And if you introduce Jev in

9:44there and because it is so quick and it

9:46is so cheap, then you'll be able to

9:48upgrade your automations through just a

9:50few prompts. And finally, we get to

9:52level three, which is building apps that

9:54have now just become possible and cost

9:56effective because of system one models

9:58like Jev. And again, this differs per

10:00person, but just to give you an idea of

10:02what I immediately use it for. In our

10:04line of work, as you might expect, I

10:06generate a lot of images as well as

10:08videos, and I put them all here in my

10:10OS, which I name as Rubric. Now, because

10:12I have hundreds of images on here, it's

10:14often the case that I need to search for

10:16specific images. And let's say if I type

10:18in claude in here, unfortunately, what

10:20this will give me are images where the

10:22file names contain the word claude. So,

10:24it's sort of like your standard CtrlF.

10:26But if we integrate Jev into that image

10:28search, and this is just a quick demo so

10:30that I can show you side by side. If I

10:32type Claude in here, you can see that

10:33the file name search here returns only a

10:36few results. But the search powered by

10:38Jev actually enables us to search images

10:40and videos by meaning instead of just

10:42the file name. And so if you have an

10:44application where users need to search

10:45for things a lot, then Jev might be good

10:47to try to see if that is going to

10:49improve user experience for your app.

10:51Another application that I found that is

10:53powered by Jev is this one from Kits who

10:56made this app called Unclutter. And

10:57basically what it does is it's a Chrome

10:59extension where whenever you toggle it

11:01on, it basically auto cleans up pages

11:03from any elements that are classified as

11:06slop. And the one that is doing that

11:08classifying is Jev under the hood. And

11:10it basically just looks at all of the

11:11elements in a page, gives a quick

11:13decision if they are ads, if they are

11:15cookie banners, and it just removes all

11:16of that when the switch is toggled. And

11:18so there you go. That is what Javis and

11:20a few use cases and ideas for you to

11:23take advantage of this new paradigm by

11:25which AI models are created and used. I

11:27hope that was useful and as usual,

11:29thanks for watching until the end. And

11:30I'm also curious like what would you use

11:32Jev for? Let me know down below and I'll

11:34see you all next time. Cheers. [music]