Full transcript
Intro
0:00Jev is the new AI model that everyone's
0:02building with right now and for good
0:04reason because it can be up to 200 times
0:06faster and 400 times cheaper than other
0:08models. And so if you know exactly where
0:10to use it and how to use it, then it can
0:12definitely make your setup cheaper, make
0:14your systems faster, and ultimately
0:16transform your workflows. So today, I'll
0:18give you 19 use cases you can do with
0:20Jev along with the prompts to get you
0:22started with your builds so that by the
0:24end of this, you'll for sure have an
0:26idea of how to best use it. We'll start
0:28with the easy ones you can try today to
0:30some more advanced ones that can truly
0:32change how you build. And by the way, if
0:34you need a quick primer on what Jeb is,
0:35I'll link a previous lesson I did
0:37somewhere in this video that goes
0:38through it in just a few minutes. And if
0:40you're new, my name is Jay. I spent over
0:42a decade working with brands you
0:43probably know, have been in AI since my
0:45masters in data science. And now I'm
0:47leading our AI business in one of the
0:48largest AI communities globally. But for
0:50now, let's dive into it. Starting with
Easy
0:52the easy ones. Number one is to put Jev
0:54in Google Sheets. One of the best ways
0:56to feel how fast Jev is is inside a
0:59spreadsheet. Jev is a decision model,
1:01which means it can't write a sentence,
1:03but it can pick true or false, a choice
1:05from a menu, or a score, and it does
1:08that in a fraction of a second for
1:10almost nothing. So, if you have data in
1:12a sheet that needs categorizing, like a
1:14bank export, you can have Claude build
1:16you a plugin where you type a new column
1:18name, say category, and Jev fills in
1:20every row while you watch. It does this
1:22really quickly and because Jev can only
1:24pick from your list, it never invents a
1:26category that you didn't ask for. Here's
1:28a prompt to get you started. Number two
1:30is to triage customer inquiries. If you
1:33get a lot of customer questions, the
1:35first job is deciding what to do with
1:37each one. Specifically, which one can an
1:39automation handle and which one needs a
1:41real person? So, what you can do is to
1:42build an automation workflow where Jev
1:44is the decision layer. Instead of paying
1:46for a big language model to read every
1:48single message, Jev tags each inquiry.
1:51Let's say if it's for billing or if it's
1:52a bug or if it's a question and it also
1:55tells you how sure it is. So that
1:57confidence score that is the useful part
1:59because if you set a rule for that
2:01confidence score like for example if Jev
2:03is less than 60% sure let's say you can
2:05just have a rule where Jev sends that
2:07straight to a person then you can just
2:08adjust that confidence number threshold
2:10as you learn what works best for your
2:12business. In one independent test, 217
2:15customer call transcripts were sorted
2:17into categories by Jev in 13 seconds for
2:20only 7 cents, which is really cheap and
2:23can really speed up your customer
2:24handling workflows. And by the way, I've
2:26put every use case in this video into
2:28this free PDF guide, which you can just
2:30send it straight to your agent and start
2:31building. The link for that is in the
2:33description below if you need it. Number
2:34three is to analyze your competitor's
2:36ads. The meta ad library shows every ad
2:39your competitors are running, but nobody
2:41reads hundreds of them by hand. So, you
2:43can have Claude build you a plug-in or a
2:45Chrome extension that does it. It saves
2:47each ad for you. And Jev reads a text
2:50very quickly on every ad and tags it.
2:52What's the format? What's the call to
2:53action? What is the funnel stage? Then
2:55everything can land in a database. So,
2:57you end up with a swipe file of what's
2:59working in your market. When I tried it,
3:01Jev tagged 54 ads from three different
3:04brands in about a second and a half for
3:06less than a cent. Number four is on
3:08clipping. If you make long videos or
3:10podcasts, finding the moments worth
3:12turning into shorts can take hours if
3:14you do it manually. So, what you can do
3:16is to build a clip finder where when I
3:17tried it out, it scored 411 possible
3:20clips from a 70minute interview in 6 and
3:231/2 seconds. Then, it also picked the
3:26top 10 that is worth posting. So you can
3:28just record once and Dev would then hand
3:30you a short list of the best moments
3:32within that long- form video. Number
3:34five is to score and profile your
3:36customer base. If you run a software as
3:38a service platform, you already have a
3:40database of your customers, how long
3:41they've been with you, whether they're
3:43on an animal plan, whether they've
3:45started to cancel, Jev can read all of
3:47that and put each customer in a group,
3:49like which ones are healthy, which ones
3:50need attention or at risk of churn. So
3:53instead of finding out when someone
3:54cancels once it's late, you already know
3:56who to reach out to this week. Number
3:58six is to build back links inside your
4:00website or your second brain system.
4:02This one is for anyone with a huge
4:04website or a big workspace where your
4:05agents are working in. If you have a
4:07site, you want your pages linking to
4:09each other because it helps people find
4:10their way around and it also helps a bit
4:12with your SEO. And if you have a second
4:14brain system, which is basically a huge
4:16folder of notes, the same links would
4:18help your agents find their way around
4:20it. When I tried it on 60 test blog
4:22posts, it took only two and a half
4:23seconds to link those pages together.
4:26And the exact same trick would also link
4:27up your notes so your second brain
4:29system would connect itself instead of
4:31assigning that to a big language model
4:33to do that for you. Number seven is to
4:35find buyers in your comments. If you
4:37sell on social, your buyers are already
4:39in your comments, but it's just buried
4:41between everything else. Take a seller
4:43account on Tik Tok for example. Every
4:45video they have brings hundreds of
4:46comments. And what you can do is to pull
4:48those comments through a scraping tool
4:50like Appify for example. Then have Jev
4:52automatically tag each one. Which ones
4:54are ready to buy? Which one is a
4:56question? Which one is a complaint? So
4:57instead of manually scrolling through
4:59your comments, you would then get a
5:00priority list with the people ready for
5:02you to buy right at the top. Number
5:04eight is about verification of Jev's
5:06results. Before we go to the rest of the
5:08use cases, one of the things you should
5:09learn is how to verify if Jev's
5:11categorization is correct for you. One
5:13way to do that is through what's called
5:15a validation set, which is just a few
5:17dozen or hundred examples depending on
5:19your use case where you already know the
5:21right answer or where a person already
5:22did the sorting by hand. So what you
5:25essentially do is you run jev on that
5:27set without showing it the answers
5:28first, then count how often it matches.
5:31You then look at where Jev got things
5:33incorrectly because that is where you
5:35can calibrate the model to be more
5:36correct next time. So, those are the
Intermediate
5:38easy ones and you can see the pattern
5:40there is basically you point Jev at a
5:41pile of stuff you already have and you
5:43just have Jev categorize those at
5:45lightning speed. Now, let's get to some
5:46of the more intermediate use cases. And
5:48by the way, if you want to learn how to
5:50build and sell AI systems that
5:51businesses actually pay for, then that's
5:53pretty much all we do over at the
5:55Robberonuggets community, where not only
5:56do you get access to the Clawed Living
5:58Master Class, which we update every week
6:00and takes you from zero to mastery with
6:02the latest on AI, but you also get
6:03access to our agents as a service
6:05course, which walks you through how to
6:07actually get paid for all these AI
6:09skills that you are learning. You also
6:10get to be part of a genuinely great
6:12community of AI builders. In fact, you
6:14can see just some of the recent wins our
6:16members are getting from the program
6:17right here. So, if you want to start
6:18earning from AI, then check that just in
6:20the pin comment below. Now, back to the
6:21video. Number nine is letting Jev pick
6:24the right skill. If you use Claude Code
6:26or another agentic platform, you
6:27probably have dozens of skills and
6:29Claude sometimes takes a lot of time to
6:31load the right one. So, what you can do
6:33is let Jev pick it. Once your prompt
6:35goes in, your skills basically becomes
6:37the menu of options and Jev would pick
6:39the right one at really fast speeds. In
6:42my own workspace, I have something like
6:44145 skills. And across 14 tests, Jev
6:47found the right skills in about 5
6:48seconds in total, while Opus 5 took
6:51around 30. The company behind Jev,
6:53TypeSafe, also tested this with 182
6:55skills, and it cut the wrong picks by
6:57more than half. Number 10 is to let Jev
7:00pick the right model. So, this is the
7:02same idea, but for model routing. So,
7:04what you can do is to make a /jv command
7:06in Claude. And when this is on, Jev
7:08would read every message that you send
7:10and pick the model for it in about a
7:12third of a second. So for example, when
7:14I ask it to find a file path, Jev handed
7:16that to a haiku helper, which is the
7:18cheapest model in Claude instead of
7:20Opus. And when the job is hard, it goes
7:22to the bigger model. So you only pay for
7:24the big model when the job actually
7:26needs it. And this same model routing
7:27idea works even if you are not using
7:29Claude. That also works in Codeex or
7:31other agentic platforms. Number 11 is to
7:34put Jev in front of your AI agents. if
7:36you have an AI agent answering your
7:38emails like these, you're paying a big
7:40model to read every single one, which
7:42includes even spam. So, what you can do
7:44is put Jev in front of it. In one test,
7:46Jev sorted 1,700 emails for 18 cents,
7:49where Jev tagged what needs a reply,
7:51what's a brand deal, and even what's a
7:53scam or a spam. Then only the ones that
7:56actually need writing or replying go to
7:58Claude, so your agent reads less, and
8:00your bill for tokens also drops. Number
8:0312 is to build Chrome extensions. You
8:05can also build Chrome extensions using
8:07Jev. And one fun example I saw is about
8:09cleaning up your X feed. Someone built
8:11one that checks every post as it loads
8:13and Jev answers one question if this is
8:16AI slop or not. If it is, the post gets
8:19folded away. So what's left is the stuff
8:21that is worth reading. The same trick
8:23works anywhere, by the way, where
8:24there's a feed like for example LinkedIn
8:26or any page. Or let's say if you want to
8:28build an extension where you want to
8:30hide the ads and cookie banners in a
8:31page like this Chrome extension called
8:33Unclutter. So you can launch plugins
8:35like these that reshape what you see in
8:37the web using Jev. Number 13 is search
8:40by meaning features. Now when you do
8:42Ctrl+F that only finds usually the exact
8:45words that you type. But with Jev you
8:47can actually build a find feature where
8:49you type roughly what you're looking for
8:50and it would highlight the right part of
8:52the page even when the words don't
8:55match. It would split the page into
8:56paragraphs, asks Jeff which one matches
8:59best and scrolls straight to it. It's
9:01fast enough to feel real time. So, if
9:03your app or your website has a search
9:05box, this is one of the easiest upgrades
9:07that you can give it. Number 14 is a
9:09related one, which is searching your
9:11images by what is in them. In my
9:13workspace, I generate a lot of images
9:15and videos, and searching them by file
9:17name barely works. So, what I built is a
9:19search powered by Jev, where typing
9:21keyword, let's say Claude, would find
9:23the images that are actually about
9:24Claude, not just the ones with Claude in
9:26the file name. The one thing you need
9:28though is some written words about each
9:30image which we call metadata. In my
9:33case, I log the prompt for every image
9:35that I generate. And so that is my
9:37metadata. But if you don't have that, a
9:39really small cheap model can write a
9:41short caption for each one. What you can
9:43try out is Gemini Flash Light because
9:45captioning a,000 images costs about only
9:4850. So those are the intermediate ones
Advanced
9:51and they already make your setup faster
9:52and cheaper. Now let's go to some of the
9:54advanced ones. Number 15, a live checker
9:57for what people say. Because Jev answers
10:00in a fraction of a second, it can check
10:02what people say while they are still
10:04saying it. Someone actually built a live
10:06BS meter called Jev meter, which scores
10:09every sentence of a debate as it's
10:11spoken. You can also build a plug-in,
10:13let's say, for your meetings where every
10:15sentence gets sorted as it is said into
10:17decisions, which ones are action items,
10:20which ones are risks, and which ones are
10:21questions. So, if you're on a sales
10:23call, you'll get easy notes to recap as
10:26you go. And it works for interviews and
10:28even podcast conversations as well.
10:31Number 16 is a chatbot with zero LLM
10:33calls. This one sounds implausible. A
10:36chatbot that never really calls a
10:37language model, but here's how it can
10:39work. In this demo app, you can ask
10:41which of our lesson teaches a certain
10:43topic. Let's say the person is curious
10:44about Jev as a topic. In the back end,
10:47Jev the model then quickly answers that
10:49in real time because it has access to a
10:51list of my videos that are already
10:53pre-transcribed, informing Jev of what
10:56each video is about. So, if you need a
10:58narrow chatbot that can answer things
11:00quickly from a given data set, like
11:02video transcripts in my case, or even
11:04company documents in most corporations,
11:06then Jev is a potential tool for that.
11:09Number 17 are smarter designs for apps.
11:12Here's a simple example for apps that
11:14rely heavily on visuals like icons. When
11:16a user types something, Jev can pick
11:19from a given set of icons. So, let's say
11:21in a habit tracker, you type drink more
11:23water and it picks the water icon for
11:25you. In a mood app, you write how your
11:27day went and it picks a custom mood
11:29icon. A designer actually showed this
11:31off where as he's typing, Jev is pulling
11:33up the icons relevant to that text. And
11:36the thing is before Jev to do this
11:37effectively you need either an LLM which
11:40can be costly or at least much costlier
11:42than Jev or to build out something like
11:43a reject based matching system which is
11:46sometimes not as effective versus a
11:48language model. But with a decision
11:50model like Jev this can be done for so
11:52much faster and for so much cheaper.
11:55Number 18 are web pages that build
11:57themselves. Now Jev can't write code but
11:59it can pick parts. Someone actually
12:01built a demo where Jev would get a
12:03library of readymade pieces like forms,
12:06buttons, signin boxes, and fonts, and it
12:08decides what each visitor needs and puts
12:11the page together on the fly. And the
12:13page shows up about as fast as a normal
12:15page load because Jev is only choosing
12:17parts and not actually building them.
12:19So, every visitor could get a page
12:21that's built for them in real time.
12:23Number 19 is about finding your own Jev
12:26use cases. This last one is a bit of a
12:28meta use case because you can actually
12:30use jev to find your own use cases.
12:32There's a site called shipwithjv.com
12:35which lists Jev use cases by category.
12:38So what you can do is to open claude,
12:40paste in that site along with a file
12:41about yourself and your business and
12:43just ask based on everything you know
12:45about me which of these use cases would
12:47help me the most. You can then ask
12:49Claude to use Jev in order to categorize
12:51which of those use case are most useful.
12:53So you get a starting point. Another
12:55option is to ask Claude to look back
12:57over your past sessions and actually
12:59point out the steps where a quick Jev
13:01call would have done the job. That way
13:03you would find use cases in the work
13:05that you are already doing. So there you
Wrap up
13:07go. That is all of them. And as a final
13:10note, OpenAI in their dev day a few days
13:12ago actually just launched their
13:14decisions API where you give it a
13:16question and a list of allowed answers
13:18and it picks one in about 150
13:20milliseconds. And the reason why that's
13:22familiar is because that's basically
13:23their answer to Jev. And the point I'm
13:25making is even if you don't end up using
13:27Jev itself for any of these use cases,
13:29getting good at this way of building
13:31makes sense now because for sure the
13:33other big labs starting with OpenAI,
13:34probably Entropic 2 in the future will
13:37be copying something like this too. As
13:39usual, thanks for watching till the end.
13:40Again, that free PDF guide with every
13:42prompt and resource from this video to
13:44get you started is just linked in the
13:45description. And uh I'm curious like
13:47which of these would you build first?
13:48And do you have other use cases in your
13:50company that I haven't covered yet? Let
13:52me know down below and I'll see you all
13:54next time. Cheers.