YT YouTube2Text

YouTube2Text → Transcripts → Claude Just DESTROYED the Reason You Pay for Software

Claude Just DESTROYED the Reason You Pay for Software — Transcript

by Corey McClain · 4,277 words · 611 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00The models are becoming the software.
0:03They're not attacking the software.
0:05They're not cloning the software. They
0:08are becoming the software. I wasn't
0:11wrong about what I said, but I was very,
0:14very short-sided because things are
0:16moving so much faster than I originally
0:19anticipated. And this new situation
0:22that's happened that's everyone's
0:23arguing about is a prime example of what
0:26I'm talking about. If you don't know
0:28what's going on, one developer prompts
0:30claw and he recreates the entire Adobe
0:33suite and then he open sources it on
0:35GitHub. It's currently at 56.8 thousand
0:39stars with 12 apps cloned in under 2
0:43weeks. There is Art Craft by Brandon
0:45Thompsons. Repos created on September
0:4830th, announced on October 3rd. It is
0:51now only 10 days later, the 13th of
0:54October. There is Photo Craft for
0:55Photoshop, Light Craft for Lightroom,
0:58Film Craft for Premiere Pro, Vector
1:00Craft, Effect Craft, PDF Craft, Design
1:03Craft. Then Word, Excel, PowerPoint, Pro
1:06Tools, and AutoCad clones on October
1:097th. It's all written in Rust and it's
1:11all written in a way so that it can run
1:13on your local machine. No subscription
1:15required. It's open source so you can
1:16look at the code. It's perfectly safe.
1:19And you can always ask Opus 5.5 to look
1:22at the code that it helped co-author to
1:24make a determination of whether it truly
1:26is safe or not before you install any of
1:28it. But I want you to see how fast these
1:30clones are moving and how fast people
1:33are actually downloading it. But it
1:36wasn't long after that that someone
1:38completely different, a week later,
1:40created a repo that decompiles any app
1:44and rebuilds the feature. And so it's
1:46called REA, Reverse Engineer Anything,
1:49and it's by Mor Luto. And it shipped out
1:52on October 7th and it hit number one on
1:54GitHub trending. It runs as an MCP
1:57server for cloud code and cursor. It
1:59decompiles, traces execution, recreates
2:02the logic. And the lesson between these
2:04two is this. The first one shows us that
2:06we can build anything from scratch with
2:08AI, but the second one proves that the
2:11binary was never hiding anything because
2:13the readme inside of REA simply tells it
2:16to look at the binary, explain the
2:19feature, and then build it yourself.
2:21It's not stealing the code and trying to
2:23violate IP. It's just learning and then
2:26recreating. And so everyone is arguing
2:29about the wrong thing. There are some
2:31people who are so big on the IP side of
2:34it. They are taking up for Adobe as if
2:36they were Adobe's lawyers and they're
2:38saying things like Adobe's lawyers are
2:40going to pound them into the ground.
2:41While at the same time, there are some
2:43people making claims that software was
2:46effectively rendered worthless. And I'm
2:49not entirely against that view. I think
2:52it's a little extreme for right now, but
2:54there is a lot of truth to it. And the
2:56ironic thing about this is that the
2:58clones own agents.md says never copy
3:02their code, shaders, or assets. Almost
3:05as though they're trying to protect what
3:07they created, but I do believe, I'm not
3:10a lawyer, but I do believe that you
3:11can't copyright code that was created by
3:14AI. The second approach that people are
3:16taking is a dismissal. Most people won't
3:19build their own software. And so they're
3:21saying things like, well, who's How many
3:24people are actually going to go out and
3:25try to recreate their own software? And
3:27I think they're right. Most people are
3:28not. But I also think that it's besides
3:30the point. But we'll come back to this
3:32in a minute. And then you have a third
3:33category of people who do not want to
3:36hear that software is dead. And they
3:38think that anyone who says software is
3:40dead has no clue of what they're talking
3:42about. But I believe that this group of
3:45people don't want to realize what is
3:48actively happening in real time.
3:50Software isn't dead. Let's just say that
3:52up front. But, there is something
3:53happening in real time that will
3:55eventually make software obsolete. But,
3:58the problem with all three of these
4:00camps is that they're all looking at the
4:01clones, but none of them are looking at
4:03what the clones are for. The models are
4:07becoming the software. They're not
4:10attacking the software. They're not
4:12cloning the software. They are becoming
4:16the software. In my opinion, AI is where
4:19the smartphone was in 2007. In 2007, the
4:23phone had a worse camera, a worse map, a
4:26worse music player, a worse calculator.
4:29Every standalone device was better at
4:33its one job. The phone was merely good
4:35enough, and it was already in your
4:37pocket. But, in 2026, a basic photo
4:40editor, a basic video editor, a basic
4:43dot tool, every standalone app is better
4:46at its one job. The model is merely good
4:49enough, and it's already in the
4:51subscription you pay for. The camera
4:53industry is what good enough looks like
4:56on a chart. I want you to look at these
4:58numbers. Smartphones wipe out decades of
5:01camera industry growth. Worldwide
5:03shipments of photo cameras by CIPA
5:05members. Film cameras are in the orange,
5:08digital cameras in the purple. This is
5:10in 1960, then the '70s, the '80s, we had
5:13a very strong uptick, the '90s, the
5:152000s, and then we see digital cameras
5:18took off, and look at how they exploded
5:21from around 30 million to as much as 120
5:25million being shipped off annually. And
5:27then we hit a sharp decline from 120
5:30million all the way down to around 20
5:33mil. And why do you think that was?
5:35Because the cameras on our smartphones
5:38became so good. They are digital cameras
5:41for all intents and purposes, and every
5:43time a new iPhone comes out, you know
5:45the camera is going to be one of the
5:47main features they talk about. There
5:48were 109 million fixed-lens cameras
5:51shipped in 2010. In 2023, there were
5:54only 1.7 million. That is a 94%
5:58decrease in total camera shipments from
6:002010 to 2023 across Canon, Nikon, Sony,
6:05Fujifilm, Panasonic, and the rest of
6:07CIPA. Now, of course, dedicated cameras
6:10still exist. I'm recording on one, but
6:12professionals buy them. market just
6:15stopped because the phone's camera was
6:17good enough, and it was already there.
6:20But, the smartphone had two things going
6:22for it. Number one, it was good enough.
6:24And number two, it was already in your
6:27pocket. You already had access to it, so
6:29you didn't have to go buy this, this,
6:30this, or that, unless you wanted that
6:32extra quality. The more capable the
6:34phone became, the fewer stand-alone
6:37items people actually bought. Now, watch
6:39that same curve start again, because
6:42it's not just landlines. The phone ate a
6:45shelf. The point-and-shoot camera
6:47dropped from 109 million to 1.7 million
6:51from 2010 to 2023. It took 13 years, but
6:55cameras eventually became something that
6:57only professionals purchased. MP3
6:59players, 54.8 million iPods were sold in
7:03fiscal 2008, the peak year. Apple
7:06stopped reporting in 2015, and the line
7:09discontinued May 2022. So, MP3 players
7:13had 7 years left. Then, landlines, about
7:165% of Americans in wireless-only homes
7:19when CDC began tracking. But, now we
7:22have 76% of US adults in wireless-only
7:25homes since late 2023. You can even look
7:28at books. Print still leads with 50.5%
7:31of US publisher revenue in 2024, but 31%
7:35of US adults read an ebook in 2025, 12%
7:39of trade sales, Q1 2025. So, books are
7:42still in there, but ebooks and other
7:44digital assets are growing. But, I want
7:47you to notice this here. The category
7:49never dies. The mass market just moves
7:52to the thing that it already owns. So,
7:55software isn't dead. Software isn't
7:56going to die. But, the mass market is
7:59going to move to the thing that it
8:00already owns. So, now I want you to look
8:02at everything that became an app that's
8:04on your phone that used to be an
8:06individual standalone item. The
8:08calculator, the scanner for PDFs and
8:11documents, pocket dictionary, alarm
8:13clock, paper maps, handheld games,
8:16flashlight, PDA, radio, voice recorder,
8:20camcorder, and even your wallet. I
8:22believe Florida and another state
8:24recently voted down allowing people to
8:25use their digital license. But, for the
8:27most part, people use Apple Pay and
8:29other type of payment systems like that
8:31on a regular basis. They save insurance
8:33cards to their phone. It's very
8:35convenient. And this is the rule you
8:37have to understand. If it's good enough
8:39and it's already there, it beats best in
8:41class and sold separately. The
8:43calculator app was never the best
8:45calculator, but it won anyway because
8:47the device you already carry with
8:50it inside of it. And you might say,
8:51"Well, you can go to the store and buy a
8:54heavy-duty scientific calculator." And
8:57And that's true. It can do all types of
8:59cosines and tangents and different
9:01things. But, most people don't even need
9:03all of that. They just need a few basic
9:06functions: multiply, add, subtract, and
9:08divide. Descript was a video editor.
9:10But, now the plan is an AI credit
9:13bucket. Like, when you sign up for
9:14Descript, and I've talked about this,
9:16they've taken some of the things that
9:17were originally included as a part of
9:20their offer, and now they shield them
9:22out based on AI credits. So, they have
9:25Underlord, the AI editor. That's the
9:27headline feature on every tier. The
9:30credits are metered. Credits meter
9:31Underlord, studio sound, green screen,
9:34eye contact, AI media, and avatars. You
9:37get all of the best voices, you can
9:39generate images.
9:41It's just become an AI platform. The
9:44editor is now the thing you spend
9:46credits inside of. It's no longer an
9:48editor. It's It's just a thing you spend
9:50credits inside of. So, you sign up for a
9:52subscription and you're basically
9:54purchasing credits to do certain things
9:56inside the platform. VidiQ, no
9:59different. They went all in on AI and
10:02kept the coaching on the back end. AI
10:04coach is now the front door. Channel
10:05audits, video reviews, competitor
10:07analysis, three thinking modes,
10:10thumbnails, clips, and short
10:11AI-generated videos come out the same
10:13chat. Media generation uses credits,
10:15which can add up quickly. Max mode is
10:17gated to the Max and coaching plans. The
10:19human coaching is the back end mode.
10:21Everyone is borrowing from the primary
10:24AI subscriptions and AI-ify other tools
10:27because the model already does what the
10:29software used to do. The only thing I
10:32might say for VidiQ is they have the
10:35YouTube API set up, so they're pulling
10:38data from multiple channels, but that's
10:39something that you can ask your monitor
10:41to do for you as well. And there's so
10:43many other platforms that are taking
10:44this approach to where AI is just
10:46becoming their everything around their
10:48business, around the service that they
10:50offer. And so, when you become a wrapper
10:52around someone else's model, the model
10:55company can skip you. You got Cloud,
10:57GPT, Gemini, and then you got the
10:59companies, the wrappers, Descript,
11:02VidiQ, and so many others. Oh, we're
11:04going to sell you credits. You get to
11:05use this user interface and work in this
11:08particular vertical. But, they're
11:10renting the capability through the API
11:12and then they're reselling it. And they
11:13add a workflow that's custom to that
11:16particular domain. But, as people begin
11:18to learn how to build their own agents
11:20and build them well and set up their own
11:21workflows and bypass all of this stuff,
11:24then the person who's the creator, who's
11:25actually using these tools, the market
11:27is going to move as this process becomes
11:30simpler and easier. Because the way it
11:32is right now, I'll admit, a good example
11:34of this is dots with Chat GPT. There's
11:37nothing a dot can do right now that an
11:39experienced or technical user of Chat
11:42GPT or Codec cannot do on their own with
11:44their own workflows. I love that
11:46platform. You can configure it to the
11:47sky, to the moon. You can do what you
11:49want with it. It's a great platform. But
11:51dots are going to be a replacement for
11:54doing all of that one day. Or for the
11:56technical users, an enhancement for
11:58doing all of that one day. When you know
12:00how to build out your own plumbing. But
12:01a dot, you just tell it what you want
12:03done, it spins up its own thread, and it
12:05takes care of the job for you. And this
12:07is great for the non-technical users
12:08because they don't they don't know how.
12:10And for a lot of them, they just they
12:12don't want to go in and do all of the
12:14extra steps. I just want to tell it what
12:16to do, and it does it. And as this
12:17process begins to become easier, you are
12:20going to find people who are not going
12:23to want the wrappers. Another example of
12:25this is Claude Motion with After
12:27Effects, released on October 8th. You
12:29can drop a prompt in, and you get an
12:31animated explainer out. Claude writes
12:33the code that animates your text,
12:35charts, shapes, and images. And if you
12:38look at the videos that Claude is
12:40generating doing this, I mean, it's not
12:42just it's not it's not 2010 clip art
12:45that Claude is working with. Claude is
12:47generating motion graphics that you
12:50would pay thousands of dollars to a
12:51designer to generate. It's generating
12:54scenes with code, videos with code that
12:56look ultra-realistic. And you can
12:58download them or open them inside of
13:01Adobe, Descript, HeyGen, Runway, Luma.
13:03Illustrations and motion effects that
13:05used to cost thousands, you can now get
13:08for a single prompt in a couple of
13:10minutes. No video model, no generated
13:13footage, every word, number, and timing
13:16is editable. So Adobe and Descript are
13:18now targets for the model's output. Even
13:21with my own workflow, I record my videos
13:24with my A-roll on my camera right here,
13:26I capture my screen for the other things
13:28that I want to share. I ask Claude to
13:30transcribe the video using Whisper
13:32locally. Then Claude goes through and
13:35plans out the motion graphics, creates
13:37the motion graphics, gives me a FCPXML
13:40that I open inside of Final Cut Pro, and
13:43then I make the final edits myself. If I
13:45need to drag something out because a
13:47word got clipped, I can do it. I can
13:48color grade sound design, and export my
13:50video. It's made my editing workflow so
13:53much easier, and it's even increased the
13:55retention of my videos overall. It's
13:57just been a win-win for me to edit with
13:59Claude because as much as I love GPT,
14:02you guys don't have taste yet. You need
14:03to work on that. Even OpenAI is a part
14:05of this trend because they shipped Pages
14:08with Sheets and Slides coming next, and
14:10Pages are a Notion killer for most
14:12people. Most of the people you have some
14:14people who are heavy Notion users, and
14:17they use every single feature of the
14:18platform. And then you have a lot of
14:20people who just can't get into Notion
14:22because it seems so difficult. Well,
14:24Pages is a very simplified version of
14:27Notion. And because it has the most
14:30basic functions of Notion, most people
14:32are going to be just fine using Pages
14:35inside of ChatGPT if they ever give it a
14:37try because they're going to realize,
14:38well, this is all I need. I don't need
14:40this full bells and whistles type thing.
14:43And there going to be people who still
14:44want the bells and whistles that Notion
14:46has to offer. And the size or the number
14:49of those people is going to reduce down
14:51to professionals in that small audience.
14:53But they also have collaborative Slides
14:55coming. They also have Sheets. And so,
14:58Spaces and Pages are going to be ultra
15:02powerful. 2009, people were saying this
15:04about the calculator. I just need the
15:06basic functions. I don't need the fancy
15:08stuff. I don't need to do coefficients
15:10and everything else with math. I don't
15:12need all I don't need a scientific
15:14calculator. I don't need it. And in
15:162026, people are saying the same thing
15:18about software. I just need the primary
15:20features. I don't need the bells and
15:22whistles. Perfect example, Descript. I
15:24just need the text editor. I just wanted
15:27to take out the bad takes. I just wanted
15:29to take out the spaces. That's it. I
15:30don't need the keep my eyes locked on
15:33the camera. I don't I don't want the
15:34green screen. That's a nice feature
15:36sometimes for some people, but I don't
15:38need the AI avatar voices. I don't need
15:40all the other things they've added over
15:42the years and and especially the things
15:44they've added since the popularity of
15:46AI. And there's just too much overlap
15:48for people not to notice what's going
15:50on. When people start saying, "I just
15:52need the primary features," which they
15:55are saying right now, that is a clear
15:57indication of where that market is
15:58headed. Software was the glue that kept
16:00people in monthly subscriptions, but the
16:02model doesn't need the glue because with
16:04software, historically, it connected
16:06data points between people. You don't
16:08know how to set up a YouTube API so that
16:10you can pull metrics from YouTube and
16:12pay for them at the same time? Don't
16:14worry. Pay VidIQ a low monthly fee and
16:17they'll pull the metrics for you. We'll
16:18even give you a great-looking interface
16:21so that you can understand the
16:22information being placed in front of you
16:24better. We'll also program the logic so
16:26everything works great for you, and
16:28we'll pull everything from the API. You
16:30don't have to do anything. It was a
16:31matter of convenience. But now, the
16:33model is the interface and the logic. It
16:36reads the binary, writes the
16:38integration, draws the UI on demand, and
16:41the middle box collapses into it. Now,
16:43it's just people, the model, and the
16:46data. That's it. And that's what REA and
16:48clones demonstrated in the same week.
16:51The middle was never the problem. The
16:53middle was never something that was too
16:56complicated for the model. The middle
16:58was always something that was
17:00vulnerable, and it was at risk of
17:02exploitation, and we are there now. And
17:04so, the objection is normal people won't
17:06clone Photoshop with Claude. And you're
17:08right, they won't. Most of them are not
17:10even going to go download the repo or
17:12the version that's free online. They're
17:14just not going to do it. Most people
17:16didn't build their own calculator app,
17:17either. But when it came out on the
17:19smartphone, I bet you they used it. I
17:21bet you if they ever needed a calculator
17:24and the thought crossed their mind to go
17:25to the store and buy one, they
17:27immediately thought, "Why? I have one on
17:28my phone." It wasn't some violent
17:30movement that made the calculator die in
17:33sales, that made digital cameras die in
17:34sales. It was just people realizing,
17:37"Oh, I have it here, right here in this
17:38convenient place with me." So, forget
17:40about Adobe costing $30, $39 a month, or
17:44$2 to $3 a day, or anything like that.
17:47Forget about the IP fight, forget about
17:49all of the cloning and the IP and
17:51different things like that. Ask what the
17:52clones are feeding. The clones aren't
17:55the product. The clones are the
17:57curriculum. The clones prove the
17:58capability gap is closed. The models
18:01that we now have can create this level
18:03of software. Now, can it create it with
18:05the fine-tuning of 26 years of software
18:07engineers? Probably not in 2 weeks, but
18:10give it some more weeks or some more
18:11months, absolutely yes. But again, it
18:14doesn't have to because most people
18:16don't need it. How many people use
18:18Photoshop? And of the people who use
18:20Photoshop, how many of them use every
18:23single feature in Photoshop? Even right
18:25now, most people are not going to
18:27download that Photoshop app because
18:28they're going to use GPT image 2.5. It's
18:31just that simple. They're going to use
18:32nano banana 3.1. It's really that
18:34simple. Like, they're not going to
18:36create images. They're going to take a
18:38photo and then ask the AI to change the
18:40lighting or change the clothes or do
18:42something else with it. Then the
18:43decompiled repo proves that the closed
18:46binaries were never a moat. 197,000
18:49stars in 6 days. Every shipped app is
18:52now readable source for an agent. And
18:55then the viral workflows are demand
18:58signal going straight to the labs. Why
19:00would Anthropic launch Claude motion?
19:03They had to have seen the people
19:06creating the motion graphics videos with
19:09Claude before Claude motion came out.
19:12Because claw has always been able to do
19:14this it just has taste but now it's an
19:16actual feature they market as a value
19:19add to your subscription and for most
19:21people what is after effects why do I
19:23need to get after effects why do I need
19:24to learn how to use this very very
19:27complex platform and spend countless
19:30hours to generate the simplest motion
19:33graphic when I can just ask my AI on my
19:36$20 subscription to do it and get it
19:38back in two or three minutes so
19:40eventually any tool that the model
19:42learns how to do better is going to the
19:44junk drawer it's just that simple but
19:46going in the junk drawer doesn't mean
19:48that it's junk and it doesn't mean that
19:50there won't be a small group of
19:51professionals who still prefer that tool
19:54there are going to be people who will
19:55build after effects by hand for the next
19:59100 years probably I don't doubt it even
20:01when AI can generate something better
20:03than them in a minute they'll still
20:05they'll still spend an entire weekend
20:07trying to get a 15 second motion graphic
20:09right and I don't blame them for it I'm
20:11just stating the facts that the market
20:13is not going to do that so right now we
20:16have docs slides design on every claw
20:19plane we have motion and dashboards in
20:22beta and dashboards are legit out of
20:25this world I really like them we have
20:27pages inside chat GPT space we have a
20:29full creative suite clone and open
20:33source by 2027 the models are going to
20:35be able to spin up software features
20:38artifacts and special tools or
20:41interfaces that you might request or
20:43that you may not request just because it
20:46thinks that's what's needed in the
20:47moment all of this stuff is going back
20:49in the training data and when it goes
20:50back into the training data and it comes
20:52out on the other side it's going to be
20:54native to the model and these are just
20:57my predictions I don't know I might be
20:58wrong but this is the direction I see
21:00things headed in cuz at the end of the
21:02day I wasn't necessarily wrong but I was
21:04so short-sighted because it's not about
21:07Jarvis, it's not about talking to your
21:10voice model and giving it commands to do
21:12things or talking to your dot and having
21:14your dot spin up threads. And it's not
21:16about Green Lantern, it's not about
21:18being able to create software. We're
21:20there right now, but that's not the end
21:22game. The end game is that the model is
21:25becoming the software. The model is
21:27becoming a universal piece of software
21:29that can shape itself into anything that
21:33you actually need at the time. And in
21:35that sense it still holds to the Green
21:37Lantern analogy, but in this case it's
21:40more like the ring itself. It can turn
21:43into anything at the given moment. And
21:45if people really understood that the
21:47model is becoming the software, they
21:49would understand that it was never about
21:51the repo, it was never about people
21:53still using Adobe, it was never about
21:55people being able to afford $30 for an
21:58Adobe subscription. It's always been
22:00about convenience. It's always been
22:02about, "Hey, I have a calculator on my
22:05smartphone. I have a flashlight on my
22:07smartphone. I have maps on my
22:09smartphone. I have a dictionary on my
22:11smartphone. I have all of these
22:13different items. I have editing tools on
22:15my smartphone if I was still doing that.
22:17I just have all of this stuff right here
22:19in my pocket. Convenience at hand. So,
22:22tell me again why I'm going to go out
22:24and purchase this other item? Again,
22:26people still buy digital cameras, but
22:29it's only a small group of
22:30professionals. But I do think that the
22:33labs might move slowly in this direction
22:36because people creating apps and coding
22:38with their models generates revenue. And
22:41of course, they could start taking big
22:43chunks out of the software industry, but
22:45if they eat too much, they will
22:46disincentivize people from using their
22:48models and stunt their own growth. So, I
22:51think it might kind of balance itself
22:52out, but over time as the models get
22:54better at just spinning this stuff up
22:56for people, I think people are going to
22:57choose convenience over proprietary
22:59software because they don't need all the
23:01bells and whistles. They just wanted to
23:03do what they needed to do in the moment.
23:04If you got value out of this video, make
23:06sure you hit the like button, subscribe
23:08to the channel, and as always, take
23:10care, have a great day, and I'll see you
23:12in this next video where I'm teaching
23:14you the new way to use ChatGPT.

About this transcript

This page contains the full transcript of Claude Just DESTROYED the Reason You Pay for Software by Corey McClain, generated from the public captions YouTube serves with the video. The transcript has 4,277 words across 611 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 translate the transcript into 100+ languages.

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, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.