Insane Jev Use Cases You Need To Use Right Now — Transcript
Full transcript
- 0:00Jev is the fastest model out there and
- 0:02people have been using it in a lot of
- 0:04insane ways ever since it was released.
- 0:06But the biggest difference we found was
- 0:08when we connected Jev with the coding
- 0:10agents that we use every day. So we
- 0:12tested Jev in our own workflow and found
- 0:15seven places where Jev makes a huge
- 0:17impact. And not only does Jev make your
- 0:19agent way faster and cheaper to run in
- 0:21these places, but it also fixes some
- 0:23problems that your agent always had. If
- 0:25this is your first time, we're a
- 0:26software company and this is our
- 0:28channel, AI Labs. And in this video,
- 0:30we're going to show you all seven Jev
- 0:32use cases, what each one changes in your
- 0:34workflow, and how to use them. This
- 0:36[snorts] section is for those who don't
- 0:38know what Jev is. And if you already
- 0:39know it, you can skip ahead to the next
- 0:41section using the timestamps in the
- 0:43description below. Now, Jev is an AI
- 0:45model made by a company called Typesafe.
- 0:47Unlike other models like Claude and GPT
- 0:50which write their answers out in text,
- 0:52Jev doesn't do that and instead it picks
- 0:54from a given set of options which is why
- 0:56it's also called a decision model. So
- 0:58instead of the way you prompt Claude and
- 1:00other models, Jev chooses from options
- 1:02whose answer can be either a yes or no
- 1:04or a list to pick from or a score to
- 1:06give to anything and it will respond by
- 1:08choosing between them. And along with
- 1:10its pick, it also gives you a number
- 1:12called confidence value. Confidence
- 1:14value is a number that shows how sure
- 1:16the model was when it chose between the
- 1:18options. And Jev is also way faster than
- 1:20other models. This is because models
- 1:22like Claude and GPT write their answer
- 1:24one token at a time. But Jev doesn't
- 1:26work like that. Jev answers all of your
- 1:29questions at the same time and gives you
- 1:31a number at much higher speed. Now you
- 1:33might think that models like Claude can
- 1:35also make these decisions themselves.
- 1:37But the problem with these models is
- 1:38that they have to think and write every
- 1:40answer. So each one costs you time and
- 1:42usage. But Jev doesn't write huge
- 1:44amounts of text like those models
- 1:46because it just gives one decision. So
- 1:48each decision takes a fraction of a
- 1:50second. Jev is also a lot cheaper than
- 1:51the usual models because it isn't
- 1:53writing tokens like the others are
- 1:55doing. You only pay for what you send to
- 1:56Jev and the answer it sends back is
- 1:58completely free. Now [snorts] Jev isn't
- 2:00a model that works inside Claude code or
- 2:03codeex as a replacement for all of the
- 2:05models that you have on it. Jev is just
- 2:07used for decision-making only because it
- 2:09doesn't write text like the other
- 2:10models. It can't be used to write code
- 2:12or let the model use tools. It can be
- 2:15used just as a decision maker inside
- 2:17tools like claude code and codeex. You
- 2:19can get access to jev via types safe's
- 2:21own platform. But the problem is that
- 2:23new signups are paused as of now because
- 2:26of the increasing demand. But it is also
- 2:28available on other platforms like open
- 2:30router and versal AI gateway. Both of
- 2:32them are platforms that give you one key
- 2:34to use AI models from lots of different
- 2:36companies and Jev is one of those
- 2:38models. You can use whichever you want,
- 2:40but we went with Versal AI gateway. To
- 2:42create the API key, you need to open
- 2:44your Versal dashboard and go to the API
- 2:47keys page inside AI gateway. When you
- 2:49create a key there, versal only shows it
- 2:51to you once, so you need to copy it
- 2:53right away because this is the exact key
- 2:55that you will be using in your workflow.
- 2:57Then you need to use the export command
- 2:59in the terminal to save the key as an
- 3:01environment variable. An environment
- 3:03variable is basically a saved value that
- 3:05anything you run in that terminal can
- 3:07read. So when you open Claude in that
- 3:09same terminal, Claude can find the key
- 3:11and use it. The first use case of Jev is
- 3:14replacing the compaction method that
- 3:16coding agents like Claude code use. The
- 3:18usual compaction basically sends the
- 3:20whole conversation back to the model and
- 3:22asks it to summarize it by following a
- 3:24set of instructions and that can take a
- 3:26long time. Jev is a good choice here
- 3:28because it can make a decision on
- 3:30whether something is important or not.
- 3:32It can actually read your conversation
- 3:34so far and decide what is important in
- 3:36the conversation and the model's
- 3:38responses. So instead of the compaction
- 3:39summary these models by default go for
- 3:42Jev gives the things that are actually
- 3:44important. To do this there is already a
- 3:46plug-in called fast Jev compaction and
- 3:48what it does is replace the compaction
- 3:50summary with Jev decisions. You can
- 3:52install it by using their install
- 3:54workflow but that repo doesn't support
- 3:56the versal AI gateway that we were using
- 3:58because it was built for Typesafe's own
- 4:00platform. So what we did was copy the
- 4:02plug-in into the project that we were
- 4:04working in. Then asked Claude Code to
- 4:06make some small changes in it so that we
- 4:08could use it with the gateway as well.
- 4:10This plug-in uses a hook to redirect
- 4:12Claude from following its usual compact
- 4:14command instructions. Since it's stored
- 4:16in the project folder rather than
- 4:17installed through the Claude
- 4:19marketplace, you'll need to ask Claude
- 4:20to update its settings. So it recognizes
- 4:23this plugin as well. Once you install
- 4:24it, you just run the compact command
- 4:26like normal. The plug-in hook takes over
- 4:28and replaces the usual compaction. It
- 4:30compacts much faster, usually in less
- 4:32than a second. There is also a limit for
- 4:34when it uses Jev for compaction. If your
- 4:36session is below 25%, it will not use
- 4:39Jev. Instead, it will use the normal
- 4:41compaction with clawed models. This
- 4:43helps make sure the Jev API is only used
- 4:46when it can make a big difference. But
- 4:47before we move on to the next use case,
- 4:50it would be great if you subscribe to
- 4:51the channel and hit the hype button.
- 4:53This small gesture of support goes a
- 4:55long way for us.
- 4:56>> [snorts]
- 4:56>> The second use case basically makes use
- 4:58of Jev's speed and low cost. And it is
- 5:00about using Jev as a checking layer in
- 5:03between that makes sure whatever you
- 5:04want is done right. When you're working
- 5:06on a project, there are a lot of docs
- 5:08and files that describe what features
- 5:09you want and what features need to be
- 5:11built. The thing is that there are a lot
- 5:13of document files which can actually get
- 5:14hard for the agent to keep track of.
- 5:16Since your default model is the one that
- 5:18worked on the feature, we can't let it
- 5:20be the judge. In order to make use of
- 5:22Jev's speed, we created a hook that runs
- 5:24whenever Claude edits the files that
- 5:26control who can access what in our app,
- 5:28and it asks Jev whether each rule in our
- 5:31docs has a test for it. This helps
- 5:33because it lets us find which rules were
- 5:34listed in the docs but never got a test.
- 5:36To use it, you just need to prompt
- 5:38Claude to check any particular part of
- 5:40your app and it will load the Jev skill,
- 5:42show you which rules are missing a test
- 5:44and then you can ask Claude to write
- 5:46tests based on the report Jev gave. The
- 5:49third [snorts] use case is about skills.
- 5:50Jev is actually a great model when it
- 5:52comes to making decisions. Agents only
- 5:54load the skills name and description in
- 5:56the context window. But there is a
- 5:58problem. When you're using a lot of
- 6:00skills on a project, the model has to
- 6:02consider all of them in the context
- 6:04window. And then from those skills, the
- 6:06model has to decide which ones to use
- 6:08and which ones not to use. That choice
- 6:10is actually something that you can make
- 6:12better by using Jev because Jev can make
- 6:14this decision much more quickly. So for
- 6:16that reason, we created a hook called
- 6:18the skill picker. This skill picker
- 6:19basically works by reading the names and
- 6:21descriptions of every skill just like
- 6:24Claude Code does and then it reads your
- 6:25prompt and picks the one skill that fits
- 6:27it or none if the prompt doesn't need a
- 6:29skill. So when you're actually
- 6:31submitting anything, it will read the
- 6:33prompt first and then tell Claude which
- 6:35skill to use for that task. So Claude
- 6:37doesn't have to work that out from every
- 6:38skill on its own. So when you're working
- 6:40on any task, you'll see that after every
- 6:42prompt, you give a hook kicks in and all
- 6:45your skills get selected way faster than
- 6:47it would have been normally without Jev.
- 6:48But before we move on, let's have a word
- 6:50by our sponsor, Zapier. If you've built
- 6:53an app, you know how quickly users start
- 6:55asking you to connect it to the other
- 6:56tools they use. What surprised us was
- 6:59how little it took. The whole
- 7:00integration is really just writing
- 7:02Node.js, and that one build reaches
- 7:04Zapier's 1 million users. We scaffolded
- 7:06a new integration from a template, then
- 7:08wired up our own app, quicklink, a
- 7:10little URL shortener we built. We
- 7:12defined its short link object once, and
- 7:14that single definition became a new link
- 7:16trigger and a shorten URL action with no
- 7:19extra wiring. We tested it locally and
- 7:21pushed it up, and it showed up right in
- 7:23the Zap editor, ready to use in a real
- 7:25Zap. And once it's in there, your app
- 7:26sits alongside Zapier's 9,000 plus app
- 7:29ecosystem, ready to connect to the tools
- 7:32people already use. and Zapier handles
- 7:34the plumbing for us from storing our API
- 7:36key to polling for new links and quietly
- 7:38skipping the duplicates. So, we only
- 7:40write the logic that's actually ours.
- 7:42Start free with the 14-day trial and the
- 7:45docs are linked in the description
- 7:46below. The fourth [snorts] use case is
- 7:48about finding files faster. Now, you
- 7:50might already know that whenever the
- 7:52model has to find a file in your
- 7:53project, it always searches the files by
- 7:56actually running an agent called
- 7:57explore. It runs in a separate context
- 7:59window and just returns the exact file.
- 8:01So it doesn't fill up the main agents
- 8:03context window with information from the
- 8:05files it opened that weren't needed. Up
- 8:07until a while ago, the default model
- 8:09Claude used for running Explore was
- 8:11Haiku, but then they changed it so
- 8:13explore now uses the same model as your
- 8:15main session. That means if you're
- 8:17working with Opus, Explore runs on Opus
- 8:192, which makes every search slower and
- 8:21more expensive. So to make use of Jev,
- 8:23you can create a skill that asks Claude
- 8:25to do what Explore does using Jev. You
- 8:27can also ask it to create an agent that
- 8:29replaces explore for this project only.
- 8:32But we chose a skill because a skill can
- 8:34also store the scripts and other things
- 8:36it needs. The skill works in two steps.
- 8:38First, it runs a normal keyword search
- 8:40to find the files that might be related
- 8:42to your question. Then Jev scores those
- 8:44files 20 at a time on how closely each
- 8:46one matches your question. And Claude
- 8:48only opens the few files at the top. So
- 8:51whenever you give it a prompt, it
- 8:52basically loads the skill and runs that
- 8:54search and ranking. And within a few
- 8:56seconds, Claude knows which files to
- 8:58read without opening all of them. Like
- 9:00it took 1.6 seconds to rank the 35 files
- 9:03and it put the right file first. The
- 9:05fifth use case is about reviewing code.
- 9:07Now you might already know that the
- 9:08agent that built a feature shouldn't be
- 9:10the one that checks it. That's why when
- 9:12we are working on our own projects, we
- 9:14always ask a different model to review
- 9:16the work. But there is a problem. This
- 9:18review that runs whole is actually slow
- 9:20because of the way Claude normally does
- 9:22step by step. So, for that reason, we
- 9:24added Jev before the review starts. Jev
- 9:26doesn't replace Claude as a reviewer.
- 9:28It's just an extra checking step that
- 9:30helps Claude review things better. Jev
- 9:32basically reads the change and answers
- 9:34seven yes or no questions about it, like
- 9:36whether this change is actually allowed
- 9:38under the rules and whether it's
- 9:39something that should be changed in the
- 9:41first place. If the answer to all of
- 9:43them is no, the reviewer only does one
- 9:45quick round. But if any answer is yes,
- 9:47or Jev isn't sure, the change gets the
- 9:49full review just like before. This helps
- 9:51because the risky changes still get
- 9:53checked properly and the small ones
- 9:55finish much faster. The [snorts] sixth
- 9:57use case is about testing your app in
- 9:59the browser. When you're working on any
- 10:00app, there are times when the agent
- 10:02opens the browser and runs the app like
- 10:04a real human would and sees if what it
- 10:06built works well or not. This helps find
- 10:08the kinds of issues that only appear
- 10:10when someone is actually using the app.
- 10:12Claude code can do this when you give it
- 10:14a browser tool and it has its model
- 10:16check whether the thing that was asked
- 10:18to be built was built properly or not.
- 10:20The model opens the browser and
- 10:22interacts with the app by clicking
- 10:23different buttons and it decides by
- 10:25itself where to click next. We can give
- 10:27that decision-making to Jev as well.
- 10:29That makes testing faster because the
- 10:31final call on whether the app works
- 10:33still comes from the main model, but Jev
- 10:35decides where to click next much faster.
- 10:37So to help with that, we asked Claude to
- 10:39create a skill for it. The skill opens
- 10:41the app and makes a list of every button
- 10:43and link on the page. And then Jev picks
- 10:45the one that gets it closer to the goal.
- 10:47It clicks that, makes a new list from
- 10:49the next page, and repeats this until
- 10:51Jev decides the goal is done or that the
- 10:53page is blocking it. In our app, we used
- 10:55it to sign in as an admin, a manager,
- 10:58and an employee and check that each of
- 11:00them can only reach the pages they're
- 11:01allowed to. And then Claude compares
- 11:03what happened against the rules in our
- 11:05docs. So, whenever you ask Claude to
- 11:07check the app in the browser, it loads
- 11:09this skill and uses Jev to test it, and
- 11:11it saves a lot of time compared to doing
- 11:13it normally.
- 11:14>> [snorts]
- 11:14>> The seventh and last use case is about
- 11:16keeping claude to your rules. When
- 11:17you've been working with agents, you
- 11:19would have noticed that there are times
- 11:20when they stop following what you
- 11:22already planned in the specs. The rules
- 11:24in your claw.md are just instructions.
- 11:26So the model follows them most of the
- 11:28time but not every time. For that, we
- 11:30created a hook because a hook runs every
- 11:32single time and the model can't skip it.
- 11:34So in our project, we have a lot of
- 11:35rules in other files that are specific
- 11:37to different parts of the app. Then
- 11:39whenever Claude is about to edit a file,
- 11:41the hook sends Jev the change along with
- 11:43the rules for that file. And Jev answers
- 11:45with a yes or no whether the change
- 11:47breaks each rule. If Jev is at least 80%
- 11:50sure a rule is broken, the edit gets
- 11:52blocked and Claude is told exactly which
- 11:54rule it broke so it can fix it. When we
- 11:56tested it, we asked Claude to create a
- 11:58file that reads the employee data
- 12:00straight from the browser, which breaks
- 12:02one of our rules. Jev decided that the
- 12:04change broke that rule, so the edit was
- 12:06blocked before the file was even
- 12:07created. and Claude told us exactly
- 12:09which rule it had broken. Now, all the
- 12:11skills and setup you saw in this video
- 12:14is available for you to use in our
- 12:15community, AIAS Pro. So, if you found
- 12:18value in what we do and want to support
- 12:20the channel, this is the best way to do
- 12:22it. The links in the description. That
- 12:24brings us to the end of this video. If
- 12:26you'd like to support the channel and
- 12:27help us keep making videos like this,
- 12:29you can do so by using the super thanks
- 12:31button below. As always, thank you for
- 12:33watching and I'll see you in the next
- 12:35one.
About this transcript
This page contains the full transcript of Insane Jev Use Cases You Need To Use Right Now by AI LABS, generated from the public captions YouTube serves with the video. The transcript has 2,806 words across 385 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.