YT YouTube2Text

YouTube2Text → Transcripts → My Jev Pi Durable Agent CURATES News: HYPE, FACT, or SIGNAL

My Jev Pi Durable Agent CURATES News: HYPE, FACT, or SIGNAL — Transcript

by IndyDevDan · 4,649 words · 668 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00What's up, engineers? Indie Dev Dan
0:01here. In the age of AI, there's already
0:04exponentially more noise than there is
0:07signal. Lots of clickbait,
0:07[clears throat] overhyped technology,
0:09and attention theft. You've likely
0:11already experienced this. I know I have.
0:13As an engineer, your time and attention
0:16are your most important resources. In
0:18this video, I want to show you how I
0:20shift through hype, facts, and signal
0:23with one ultra-important emerging piece
0:26of technology, proactive agents. But the
0:29meta lesson of what you'll see in this
0:31video is more important than the tools
0:34we'll use. The best engineers are
0:36transitioning from human in the loop to
0:38human near the loop to human out the
0:40loop. In, near, out. Why is this
0:42transition so important? It's because
0:45your agents can do 80% or more of
0:48everything you're doing right now
0:50autonomously [music]
0:51if you have the right tech in place. The
0:54problem is jumping from in loop straight
0:56to [music] out loop is a big challenge.
0:59That's where proactive agents come in.
1:01Proactive agents don't wait for you
1:03[music] to work. They work for you, then
1:05wait. And when you're ready, you step
1:07in, extract value, [music] give
1:09feedback, and move on. If you want to
1:11understand how you can up-level your
1:13agentic engineering by moving from
1:15[music] in, near, and out the loop,
1:17stick around and let's filter out the
1:19noise with the durable news agent.
1:27What if your agents never stop?
1:34So, let's pull up the application, and
1:36what you'll see here is a typical agent
1:38on the right, generative UI on the left.
1:41You can see here, our news agent has
1:43just caught the recent full announcement
1:46of Gemini for Argon getting released via
1:49their APIs. So, this is very powerful,
1:51but at first glance, you might think
1:53this isn't super special. Might just
1:54recreating ChatGPT here. Not so fast. We
1:57can access this agent in two ways. You
1:59could see here I'm not actually
2:00operating on my device. I'm actually
2:03looking at a URL on my M4 Mac mini
2:06device here. That means that the browser
2:08we're looking at is a detached client.
2:10Let's push this further. This is what a
2:12proactive agent really is. It's not
2:14waiting for me to work. If you open up
2:16my terminal, you can see something
2:17really amazing here. My agent in the
2:19terminal and in the browser, they are
2:21synced. Why are they synced? It's
2:23because these are both only clients. My
2:25real agent is operating on my M4 Mac
2:28mini. The browser and my terminal user
2:30interface here, they can see are both
2:32running GPT-6 Luna, are operating and
2:35connected to my durable agent running
2:38directly on the M4. So that's the power
2:41of what we have here. And you can see,
2:43you know, my hands are off the keyboard.
2:44Every minute here, every cron interval,
2:47work is just going to happen. It's not
2:49waiting for me to work. It's working for
2:51me and then waiting for the next
2:53execution to kick off. This agent is
2:55proactive. It's built into a system.
2:58It's more like a product that's
3:00operating without me, for me. It really
3:02is that simple. A proactive agent is a
3:05durable agent plus a trigger inside of a
3:07sandbox. We can quickly just spam
3:09refresh here and get all the latest
3:11information. This agent is durable. It's
3:13backed by SQLite databases. It's backed
3:15by any database you really want to. What
3:17we're operating now is just a client.
3:20It's a thin interface to an API which
3:23has the actual agent that's durable
3:25inside of it. This was the topic of last
3:27week's video where we looked at an
3:29on-call durable agent that was also
3:31proactive. And guess what? We're focused
3:33on it here again because this is a
3:35high-signal agentic piece of technology
3:38that's critical for advancing from in
3:40the loop, which is what this is, to near
3:41the loop, which is what this is, to
3:43fully out the loop. Thanks to
3:45intelligence going up and cost going
3:47down. I've had this running for over a
3:49day now. Just non-stop gathering news,
3:52filtering it out, using Jev as a key
3:55part of the pipeline for gating the more
3:57expensive language model calls. Even
4:00though this is a very cheap Luna model,
4:02it's still a language model call that's
4:04chewing up tokens, compute, and
4:06therefore cash. So, we put Jev in front
4:08to gate, to filter, to look for the
4:10right pieces of information. And this
4:12allows us to smoothly transition from in
4:15loop to out loop. The system is working
4:17for me without me. So, pretty standard
4:19UI. This is a standard modern user
4:21interface that you're probably used to.
4:22We have agent on the right or on the
4:24left, and then we have the generative UI
4:26in the other portion. You can see we
4:27have a nice timeline here. We'll break
4:29down the actual feature set here because
4:31how something is built matters more than
4:33what you're building. Because as you
4:35know, code is now infinitely cheap,
4:37infinitely copyable. The reverse
4:38engineer anything trend is showing that.
4:41Mathematicians losing their mind over
4:43OpenAI's huge math proof dump is also
4:46showing that. What matters now is your
4:48taste, your judgment, and your ability
4:50to architect unique systems with unique
4:52properties. So, before we get to the
4:54architecture of this durable news agent,
4:56which looks for news that's actually
4:58worth your attention. You can see we're
5:00getting a lot of news on the Gemini 4
5:02Argon release. That's the fresh new hot
5:04thing coming in right now. And my agent
5:06pulled this to me. I didn't have to go
5:08find it. But before we dive into the
5:09architecture, I just want to talk about
5:10like, how did I apply my taste, my
5:12expertise to drive what I built here?
5:19Mentally, I separate new information
5:21into three buckets: hype, facts, and
5:24signal. Hype usually means there's too
5:26much attention and topic growth is
5:28driven by unfounded excitement because
5:30the topic itself is arousing or
5:32attention-grabbing by nature. So, this
5:34is hype. We then have facts. And facts
5:37are objectively true. This is much
5:39closer to what I call signal, but it's
5:41important to differentiate signal from
5:43fact because not all facts are useful
5:45information you can actually act on. For
5:48instance, OpenAI dropped that huge set
5:50of math proofs. Can you, an engineer,
5:52actually act on that? Not really, but it
5:54is still awesome. It is still a fact of
5:56the matter. There's a new wave of
5:58mathematics coming that will likely
6:00unlock things in the future for you and
6:02I and for the world. This is a fact
6:04though. It's not really signal right
6:06now. And then there's signal. Signal is
6:08something you can actually act on. For
6:10instance, Gemini 4 Argon is a new
6:12language model. It's state of the art.
6:13You and I can use this right now. It was
6:16announced like a week ago or something,
6:17but this has just been rolled out at the
6:19time of filming. And now this is an
6:21actual piece of information you and I
6:22can act on. So, this is how I like to
6:24separate information in the age of AI to
6:26filter out the noise. So, with this
6:28applied taste, aka judgment, I can now
6:31build an architecture to help me do
6:32that. And obviously you're seeing the
6:33end result here, but I want to walk you
6:35through why software engineering is
6:38still so important and having an
6:39opinion, having a product vision still
6:42matters. In fact, it matters more than
6:44ever since everyone is now copying,
6:46reverse engineering, and slopping out
6:48anything they can think of. What does
6:50the architecture of this look like?
6:51Because now we have a goal, now we have
6:53a concrete opinion about what the result
6:55should look like. Now we can dive in and
6:56answer like, what's the architecture?
7:02So, the architecture for the system
7:04looks like this. This is the whole
7:05system. We have cron firing off our API
7:08and our API is our news desk Pi Durable
7:11Agent. We dove into the details of Pi
7:14Durable Agent in our last week's video.
7:16That one's getting a lot of attention
7:17for good reason. I think a lot of
7:18engineers are understanding this concept
7:20of proactive agents. It's so important.
7:22I have to talk about it again
7:24back-to-back week after week. This is
7:26high signal. Pi Durable lets us create
7:29durable tasks. This is all triggered via
7:31an API. So, everything is isolated. And
7:33again, this is my Mac mini device. I can
7:35deploy this anywhere. In fact, I have.
7:38If we jump back, you can see this is
7:39operating on a local URL, but this one
7:42here is operating on a exc.dev agent
7:45sandbox. As you can see, this one's
7:47focused on muse. These are different
7:48sets of information. It's asking me some
7:50questions. The system is prompting me.
7:53This is what's called a two-way prompt
7:54system. I can prompt it, it can prompt
7:56me, and it's creating this own unique
7:58feed that's completely separate from my
8:00feed on my M4, which is more fine-tuned.
8:02So, you can see I'm testing on my M4 and
8:04on an exc.dev sandbox. So, again,
8:06detached clients is the theme here with
8:08durable agents. And this thing is just
8:10going to keep running, as you can see
8:12here. And as you saw, we have the
8:14terminal user interface. We can access
8:15this via CLI, and we can access it via a
8:18browser. So, this is the whole system.
8:20It's a client-server application, where
8:22we've separated out several of the
8:24moving parts. So, what does that look
8:26like?
8:30This is built in layers. So, we have our
8:32SQLite databases. We have Pi Durable.
8:35Shoutout Mario and Armin. We then have a
8:37durable agent. This is a custom
8:38implementation of it, which then is
8:40exposed via an API. We also have Jev,
8:43which is doing a lot of classification,
8:44a lot of filtering, a lot of ranking at
8:46dirt-cheap prices. And then we have the
8:48actual SDK, which our browser, our TUI,
8:50our cron, and our CLI uses. What we're
8:53doing here is building an architecture
8:55that can be redeployed across many
8:57different use cases in a proactive
8:59agent-first way. That's the idea.
9:01Whenever I sit down to build something,
9:02it's not just about the thing. It's
9:04about how it's built. It's about the
9:06architecture around it. There's more and
9:08more fear, uncertainty, and doubt in the
9:10age of AI. More than ever now, code
9:12generation is solved. People are saying
9:14mathematics is solved. People are saying
9:16software engineering is solved.
9:18>> [laughter]
9:18>> I completely disagree with this. We're
9:20still going to need great practitioners
9:23who put their hands on the keyboard to
9:25drive the right results, to use the
9:27right architecture, to ask the
9:28questions, and to guide the language
9:31models. Obviously, there's levels to
9:32that. That's what we're always focused
9:34on here on the Endy Dev Dan channel, but
9:35like, don't lose faith in your
9:37engineering skills. You've been doing
9:38this for years. Many of you have been
9:40engineering like I have. I've been
9:42engineering for over 15 years. And
9:44agents have now come onto the scene.
9:45This is a new tool. That's all it is.
9:47Use it as a new tool. Don't let it scare
9:49the out of you and prevent you from
9:51actually getting things done because you
9:53can do more than ever now if you're
9:55using the right tools. So, you can see
9:56here we're coming in with some dead
9:58sweeps. This thing runs every single
9:59minute, so there's a lot it's pulling
10:01in. But, back to the architecture. It's
10:03built in layers and this unlocks this
10:05type of system. This is how this works.
10:07This is the AI developer workflow. On
10:09this top-level new sweep, we then have a
10:10task that runs. We gather, we gate, we
10:12dedupe, and then we rank. And then we
10:14have n candidates that come back, and
10:16then our durable agent decides what to
10:19actually render based on the rankings
10:21that come back. So, that's the
10:22architecture. That's how it works. And
10:24we're pretty much using Jev at every
10:25point in this.
10:29This is a great way to classify large
10:31amounts of text in many, many different
10:33ways. And Jev is, of course, as you
10:35know, dirt cheap. It's super reliable.
10:37It's very fast. Roll your own if you
10:39want. I am saving a crap ton of time by
10:41just using out of the box Jev. It's a
10:44great generic classifier. But, you can
10:45see here, this is a very powerful way to
10:48process large amounts of information at
10:51sub half-second times per call. You can
10:54see how we're using this here. Is this
10:55news piece relevant? Is it a dupe? Is it
10:57the same story? If you can't hash it,
10:59this is a great way to fall back from
11:01hashes. Of course, you want to do a
11:02content hash first. Then, of course, you
11:04know, research claim. Is this actually
11:06fact? We can put together many sources
11:08and determine if there's some overlap.
11:10And then we do actual presentation. The
11:12big idea here, the meta lesson, the meta
11:14idea behind what I'm showing you here is
11:16this.
11:20We want to move from in loop to near
11:22loop to out loop. This is the name of
11:24the game for agentic engineers because
11:26this is a measure of your skill. It's
11:28not about sitting in the terminal
11:29prompting back and forth. You've heard
11:31me say that a million times now. The
11:32real edge now is move from in to near to
11:36out. And this is a direct measure of
11:39your agentic engineering skill. Because
11:40what does it mean? It means that you
11:42have more autonomous systems working for
11:44you all the time, doing real work for
11:45you all the time. Of course, you know,
11:47the tokens you use have to be useful.
11:49The code you write, the things that run
11:51have to be useful. As you do this, as
11:53you make this transition, which is what
11:54it's really all about, this is what it
11:57looks like to increase your agentic
11:58engineering skill. It's more threads of
12:00work working for you on your behalf all
12:03the time, around the clock. This is the
12:04new mental model I'm using and
12:06recommending to engineers in the loop,
12:07near the loop, and then out the loop
12:09completely. And I hope you can see how
12:11this system is a direct replication of
12:13that. First, you're in, then you're
12:15near, and then you're out. Let me talk
12:17about some of the actual feature sets of
12:20the durable news agent. Let's move from
12:22engineering and architecture to hands-on
12:24[music] product.
12:29So, what are the features of this tool?
12:30One key thing everyone should be doing
12:32in their software now is creating a data
12:34flywheel. It's creating a feedback loop
12:36between your users' inputs and the way
12:38your code plus agents takes that
12:40feedback and improves the product. News
12:43and gathering information and filtering
12:45through the noise is a great example for
12:46this. There's a reason I'm targeting
12:48this and I'm improving my agentic
12:50engineering by using this use case
12:51specifically. It's because I can come in
12:53now and upvote or downvote something.
12:55This is real information. This is real
12:57signal and fact that I care about. First
12:58off, I can just thumbs it up, and then I
13:00can upvote the fact that this model
13:03release has happened. Of course, if I
13:04click this, it'll open up the full link
13:06and we can read it and dive into it. We
13:07can see how this is getting trust me,
13:08bro, benchmarked right here, and I'm
13:10sure the actual post goes a lot further.
13:13If we check out the Google DeepMind
13:14official release from 35 minutes ago, we
13:17can go ahead and dive into this and see
13:19the actual news on this release. I'm
13:21filming this at the perfect time. Our
13:23ProActivation picked this up and is now
13:25filling the news feed with this. It's
13:26fantastic. We'll see if that model
13:28actually gives us anything unique, but
13:30this is signal that I care about. You
13:31can see that our Jev classifier said,
13:34"Yep, this is mostly signal. There's a
13:35lot of fact here. It's a new model
13:37release."
13:38There's nothing to argue there. And then
13:40some hype, which is kind of true, 66
13:42hype out of 100 because these model labs
13:44always have to hype up their release.
13:45So, you know, this makes a lot of sense.
13:47So, that's one of the key features I can
13:48thumbs this up and I can reinforce back
13:51into the ranking algorithms that this is
13:53important. Let me do this. I'll pass
13:55this in and I'll say increase the weight
13:58of Gemini. If it doesn't exist as a
14:01keyword, add it. So, I'm going to talk
14:03directly to my agent. I'm going to queue
14:04this message. It's completing its sweep,
14:06so my message should pop up here in just
14:09a moment. Check. There we go. Good. And
14:11there's my message. I can quickly
14:13reference that card. Increase the weight
14:14of Gemini. If it doesn't exist, add a
14:16keyword for it. And so, it looks like
14:17there is a task queuing here. There we
14:19go. It's reading my taste files. This is
14:22all the settings. So, in the settings
14:23here, you can see we have all the
14:25keywords that I'm focused on and we have
14:27the links that our system normally uses.
14:30This is all available here to be tuned
14:31and customized. Pulling in information
14:33is all about weighting it. What do you
14:35care about the most? The system lets me
14:37weight, increase, decrease the signal
14:39for things. You can see here my agent
14:41edited that keyword. It raised the
14:42existing Gemini keyword topic from 1.2
14:45to 1.3. So, in all subsequent runs, it
14:47knows that this is an important keyword
14:49that I personally care about. So, what
14:51is this application really doing here,
14:53right? This is a high signal news feed
14:55curated to my taste driven by a feedback
14:58loop between myself and my durable
15:00agents, specifically my durable agent
15:02architecture. So, you can see here we
15:04got one new piece of news sent in. And
15:06we can see this is banking watchdogs
15:08ramp up AI scrutiny. This is some
15:09signal, some fact. If I want to, I can
15:11get rid of this. There's maybe some
15:13important information here. I'm just
15:14going to go ahead and downvote that, get
15:15that out of my stream. But this is a
15:16really important feature. I can increase
15:18and decrease the signal arbitrarily of
15:21what matters. Your signal, what you care
15:23about, needs to be fine-tuned. You can
15:25see a bunch of things that I mentioned
15:27here. All these are, you know, if you've
15:28been with the channel, they're all
15:29obvious topics and keywords that I focus
15:32on a lot on the channel. We also have
15:34links, news sources. I'm just starting
15:36to to tweak and tune the weights of my
15:38actual pages. Shout out Hacker News,
15:41Latent Space, Pi, Cloud Code, Anthropic
15:43Codex, OpenAI, Simon Willison. And this
15:46is all fueled by an agent that has a
15:48unique set of tools. So, this is
15:50important for a lot of engineers really
15:52stuck in the engineering world. You
15:53should be building products with agents
15:55with unique actions. They should be able
15:57to do unique things that differentiate
15:59your product. This is a durable news
16:01agent. All it does, it's not reading
16:03files, it's not changing code. It has
16:05specific tools to operate and to filter
16:07and to curate news. So, we have run the
16:10news sweep, list candidates, read items,
16:12search web, scrape URL, research items,
16:14Jev classify. Huge shout out to Jev and
16:16the Type Safe team. Present news,
16:18present timeline, ask taste, so on and
16:20so forth. All harness engineered tools
16:22that I built with an N loop agent to get
16:25the system up and running. I'm building
16:27the system that builds a system. That's
16:28the reason you go into the loop. You're
16:30working on a system that's going to work
16:32for you with less of your attention. You
16:34want to be near the loop or out the loop
16:36and spending less time in the loop.
16:38That's the big meta theme. So, we have a
16:40nice set of tools here and we've got two
16:41new news blocks that just came in. Two
16:43nice new Hacker News drops. And you can
16:46see the keyword on the left. We've got
16:47coding agents and agentic engineering.
16:49This looks great. And the agent is just
16:50going to keep running. This tick is on
16:52every minute. It's going to look for a
16:53maximum seven cards. It's going to Jev
16:55rank them. So, you can see, you know,
16:57this system is very very powerful. It's
16:58a great application. The durable agent,
17:01specifically I'm using Pi Durable as
17:03mentioned in last week's video. If you
17:04want to learn more about their specific
17:06implementation, which is what I'm using
17:07under the hood. Big shout out to the Pi
17:09agent, my favorite open source agentic
17:11coding tool. Because guess what? I'm in
17:13control. I can customize it and I can
17:15extend it however I want. It's the same
17:16with their durable agent implementation.
17:19And this is the new mental shift that I
17:20think durable agents unlock. They allow
17:22proactive agents that really create this
17:25middle step for you. It makes it easier
17:27to transition from in loop to out loop,
17:30which is the central idea behind the
17:32phase three [music] agentic engineering
17:34product I'm building right now for
17:36advanced engineers using agents every
17:38single [music] day. Link in the
17:39description for that. Early sign up is
17:41available. Don't miss that because this
17:43next leap moving to out loop is truly a
17:46compounding, dare I say exponential
17:48opportunity, an exponential leap in your
17:50engineering capability. As you can kind
17:52of see, there's hints of it already by
17:54just understanding and internalizing
17:56near the loop agentic engineering.
17:57First, you're in the loop, then you're
17:59near the loop, so you're spending less
18:00time, and then you're completely out the
18:02loop. And at every level, your ability
18:04to engineer, this is not vibing, this is
18:07not vibe slapping, your ability to
18:08agentic engineer the system becomes a
18:10higher and higher requirement to
18:12actually build something useful here
18:14that can work without you. Same for near
18:16the loop. So, this is a new mental
18:17framework, the new mental model I'm
18:19recommending to engineers, and you know,
18:20this is the way forward because this is
18:22going to offer you the most leverage,
18:24but you have to make the transition in,
18:26near, out.
18:30The lower level core theme here is
18:33durable agents are powerful. This
18:35durable news agent and last week's on
18:37call incident response agent, they're
18:38both really concrete examples of what it
18:41looks like to build an agentic system in
18:43a really concrete way with a concrete
18:45outcomes that operate for you without
18:48you. And so, you can see the system will
18:49just continue. We've got all the
18:50executions and all the traces mapped out
18:53that we can dive into at any point in
18:54time. We've got a nice set of settings
18:57with weighted keywords that we can up
19:00and down vote. The durable agent is on
19:02my m4, it's on my exe.dev cloud sandbox.
19:05Great, we've got a nice jab hit here.
19:08I'm going to be tuning and tweaking this
19:09algorithm. Linked in the description for
19:11you is going to be this durable news
19:13agent system. And the whole point here
19:15is to really help you fast-track that
19:17movement between in the loop, near the
19:19loop, and out the loop. So, take this,
19:21tweak it, make it your own. I'm not
19:22going to give you
19:23>> [laughter]
19:24>> my settings. You're going to have to
19:25choose your own settings. Feel free to,
19:26you know, take a snapshot, try to get it
19:28into the system as much as you want here
19:29with some of the settings I'm playing
19:30with, but there are a lot of
19:31improvements we made here. I'm really
19:33happy with the early implementations of
19:34this. I'm going to be continuing to
19:36build on this. Again, link in the
19:37description. You want to be thinking
19:39about how you can build agents that work
19:41for you without you. Proactive agents
19:43don't wait for you to work. They work
19:44for you, then wait. On the channel,
19:46we're constantly pushing toward out-loop
19:49agentic engineering. If you haven't
19:50already, definitely sign up for the
19:52phase three agentic engineering product
19:55I'm building right now. Hands-on, this
19:57is for the mid to senior correct agentic
20:00engineer that's on the edge. Here, we're
20:02focused purely on out-loop phase three
20:05agentic engineering. All phase two
20:07tactical agentic coding members will
20:09receive a discount. So, if you're
20:11interested in that, check it out. That's
20:12phase two agentic coding. Here, we're
20:14going to be focused purely on the next
20:16phase, the next massive leap. Pre-email
20:19sign-up is available now. In a couple
20:21weeks, it'll be pre-sell. And at the end
20:23of November, it'll be launch time. It'll
20:25be go time. I can't wait to share more
20:27of this with you. Link in the
20:28description. Get signed up. The next
20:30leap is going to be massive because once
20:32your out-loop system starts compounding
20:34for you, starts working for you, it's
20:36not going to stop. And durable agents
20:38are a key step in that direction. In AI
20:403, we're going to take that much
20:42further. I've got the durable news agent
20:44linked for you in the description. Check
20:46this out, make it your own, tweak it,
20:47add more API providers, add more links
20:50so that you're filtering out all the
20:52junk coming toward you in the age of AI
20:55to just the most important pieces. You
20:57can see here, I've got a decent set of
20:58information coming my way, but lots to
21:00improve, lots to tweak here. Costs have
21:02barely moved. I'm paying close attention
21:04to the tokenomics of the system so that
21:06I don't go broke in generating the
21:08value. Jev and cheap language models
21:10like Luna and Haiku 5.5 are going to be
21:13key in that, as well as other models
21:14like Deep Seek V4.1 Flash. I highly
21:17recommend you start transitioning from
21:19in-loop to near-loop to out-loop. You
21:21can see my terminal client still
21:23completely synced up. We can say what
21:25was the last entry you added, and it's
21:27going to be completely synced with the
21:29browser. It's currently queued till
21:31after the sweep runs. There we go. What
21:33was the last entry you added? We got a
21:35new post from Anthropic saying code
21:37merges have rose nearly tenfold since
21:40May. It got correctly categorized as
21:42hype. Also some fact, some signal, but
21:44mostly hype. I completely agree with
21:46that. Thumbs up on that. The models
21:47continue. If you want to learn more
21:49about productive agents, check out the
21:50link in the description to check out
21:52last week's video where we use Py
21:54Durable to build an on-call incident
21:57response agent. Productive agents are
21:59the future of engineering. You want
22:00agents that are working for you [music]
22:01in the background and waiting for your
22:03feedback, creating a data flywheel for
22:05both you and your users of [music] your
22:07products. I'm building out more and more
22:09of these all over my stacks, all over my
22:11devices, in the cloud, in [music]
22:12products, because the signal here is
22:15high. The signal here is rich. I
22:17recommend you do the same. Definitely
22:18sign up for the phase three product if
22:20you haven't, and keep your eye on the
22:21signal. [music] If you made it to the
22:23end, let the YouTube algorithm know this
22:25is signal you care about. [music] You
22:26know where to find me every single
22:28Monday. Stay focused and keep building.

About this transcript

This page contains the full transcript of My Jev Pi Durable Agent CURATES News: HYPE, FACT, or SIGNAL by IndyDevDan, generated from the public captions YouTube serves with the video. The transcript has 4,649 words across 668 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.