Reverse Engineer Anything Does Way More Than You Think — Transcript
Full transcript
- 0:00Reverse engineer anything just went
- 0:01viral and people are freaking out
- 0:03because it's a free tool that lets you
- 0:04rebuild almost any app in minutes. Now,
- 0:07I will show you how to use this, but
- 0:08that's not the point video. Instead, I'm
- 0:10going to give you a step-by-step
- 0:11playbook to reverse engineer anything
- 0:14including non-technical work. So first
- 0:16we'll cover what REA is and use it to
- 0:18rebuild an app. Then we'll apply the
- 0:20exact same playbook to reverse engineer
- 0:21four things, successful businesses,
- 0:23specific tools, tasks you want to
- 0:25automate, and strategic thinking from
- 0:27industry experts. I guarantee whether
- 0:29you're technical or not, this will
- 0:30change how you use AI every single day.
- 0:32So what REA is and how it works. So REA,
- 0:35short for reverse engineer anything, is
- 0:37an open source tool that anyone can use
- 0:39to rebuild a computer app, but it's
- 0:41actually nothing new. It's really just a
- 0:43combination of several pre-existing
- 0:45tools that engineers have used for
- 0:47years. So Ghidra, Hopper, and IDA, which
- 0:50take apart compiled software, Chrome Dev
- 0:52Tools, which show you what a website is
- 0:54doing, and app code reader, which walks
- 0:56through an app. And these aren't new,
- 0:57right? Some of them have actually been
- 0:58around since the 1970s. So what makes
- 1:01REA as unique is it's a harness. It lets
- 1:03any person talk to all of these tools at
- 1:05once without knowing how to use any of
- 1:07them. On screen you can see a diagram
- 1:09directly from REA's own documentation
- 1:11and it's essentially just a loop. First
- 1:13you give your AI agent a target, for
- 1:15example an app. Then REA inspects it and
- 1:18traces it with its analysis tools. Third
- 1:20the agent then takes that information
- 1:21and then starts building the app. It
- 1:23then keeps looping until the app is
- 1:24entirely recreated. So there is a bigger
- 1:26story here about what this means for the
- 1:28future of software and I'll get to that
- 1:30at the end of the video. But first, how
- 1:31do you actually use this thing? So first
- 1:33download REA by running this prompt.
- 1:35Then you can pick the app that you
- 1:36actually want to recreate. On screen I
- 1:38chose to reverse engineer Apple Notes
- 1:40and after about 40 minutes here is the
- 1:41result. On the left is Apple's version
- 1:43and on the right is the recreated one.
- 1:45They're pretty damn similar and it is
- 1:47awesome, but again that is not the point
- 1:49of this video. You're likely a business
- 1:51owner or a high-level operator watching
- 1:52this channel. How can you actually apply
- 1:54this to get value from it? Just ripping
- 1:56off a notes app is genuinely useless. So
- 2:00what you actually take from this? Well,
- 2:02REA has put a spotlight on something
- 2:04people have been doing forever, and
- 2:06that's figuring out exactly how
- 2:08something works, then reverse
- 2:09engineering it to rebuild it. And in
- 2:12every field, not just technical ones,
- 2:13people have been doing this. So for
- 2:15example, writers. They learn by studying
- 2:17authors they admire. They take apart the
- 2:19sentences, borrow techniques, and
- 2:21eventually they find out their own
- 2:22voice. Now, if you're not a writer,
- 2:24let's say in business, it's the same
- 2:26process. We borrow pieces of successful
- 2:28businesses and their processes to make
- 2:29them our own. So here is how you can
- 2:31actually do this yourself. REA works in
- 2:33five steps, and we're going to borrow
- 2:35them for the rest of the video. First,
- 2:37pick what you're studying. Second,
- 2:39gather the evidence. Third, design what
- 2:41you'll build from it. Fourth, rebuild
- 2:43and grade it against the original. Five,
- 2:45iterate yourself and make it yours.
- 2:47That's the process to learn anything
- 2:49from winners and make yourself more
- 2:50productive. So the first thing we're
- 2:52going to reverse engineer is businesses.
- 2:54There's a famous process Russell Brunson
- 2:55popularized called funnel hacking. You
- 2:58walk through a competitor's funnel,
- 2:59which is how they turn strangers into
- 3:00customers, and study the exact steps
- 3:02they use. And we're going to systematize
- 3:04that using our five-step REA framework
- 3:06so you can see what makes them so
- 3:08successful. So first, pick the business
- 3:10you want to analyze. And a general rule
- 3:12of thumb for whatever business you're
- 3:14analyzing, the bigger the player, the
- 3:16less you should study them. And that's
- 3:18because let's say you're looking at
- 3:19Nike, their brand props up whatever work
- 3:21they're doing, so you can't just redo
- 3:23what Nike's doing because you don't have
- 3:24that brand authority. So instead, pick
- 3:26the small competitor that just broke out
- 3:29and see what they're doing right. So use
- 3:30this prompt to analyze and identify your
- 3:32top competitors. And no, you don't need
- 3:34to run a business for this to be
- 3:35valuable. For example, if you're in
- 3:36marketing, see what your marketing
- 3:38competitors are doing. Second step,
- 3:40gather data you can leverage. Run this
- 3:42prompt to search through a competitor's
- 3:44page for info. And you'll find that some
- 3:46websites block automated tools, so you
- 3:48won't be able to see their subpages in
- 3:50the funnel. And so that's where you have
- 3:51a couple of options. One, you can have
- 3:52Claude walk through it in your browser.
- 3:54And the key here, whenever you do this,
- 3:56I would tell Claude to take screenshots
- 3:58of any key steps. In my eyes, think of
- 4:00these screenshots as your evidence pile.
- 4:02They remove the interpretation gap
- 4:04because you can see what Claude saw
- 4:06instead of trusting its summary. But at
- 4:08this point, you may still get blocked
- 4:10and this is very slow. So if you want a
- 4:11faster, more effective way to do it,
- 4:13then effectively look through each page
- 4:15and get through any blockers. I added a
- 4:17competitor research tool to
- 4:18buildpartner.ai, which you can use
- 4:20entirely for free. And you can run this
- 4:21single command and it'll go through the
- 4:23funnel and do a deep analysis. Step
- 4:25three, design. In this case, that means
- 4:27taking what you learned and applying it
- 4:29to your work. Here's a prompt that takes
- 4:31that information and gives you
- 4:32actionable takeaways for your business.
- 4:34Four is rebuild and grade. Take that
- 4:36list from the previous step and have
- 4:37Claude build what it can, then evaluate
- 4:39the final product. And finally, step
- 4:42five is iterate. See what's working, see
- 4:44what didn't, see what other people are
- 4:45doing over time. This is a continuous
- 4:47learning process. Now, as we go through
- 4:49each of these use cases, both REA as
- 4:51well as anything that we're going to
- 4:52cover, there are limitations. So neither
- 4:55this funnel hack nor REA can see the
- 4:57back end of how things work. So in the
- 4:58business example, it can't see the
- 5:00customer support, [music] the email
- 5:01follow-ups, etc. And for REA and the
- 5:03computer example, it can't see anything
- 5:05that it doesn't have direct access to.
- 5:06So for example, the back end or the
- 5:08database of the tool itself. But it
- 5:10doesn't have to be that extreme.
- 5:11Sometimes the most valuable thing is a
- 5:13single tool that you recreate. But
- 5:15before we get to number two, building
- 5:16tools, everything in this section was
- 5:18about studying how someone else turns
- 5:19strangers into customers. What it can't
- 5:21show you is where those strangers come
- 5:23from the first place, which brings us to
- 5:25today's video sponsor, Clay. So everyone
- 5:27on this channel is learning to build
- 5:28with AI, but nobody really talks about
- 5:29who you're actually building it for. And
- 5:31finding customers is still the hardest
- 5:33part. Now, there are two ways to
- 5:34actually get these customers. First, you
- 5:36can build an audience, which takes
- 5:37years, or go find the right people and
- 5:40reach out to them directly. And you can
- 5:41start doing that today. And that's what
- 5:43Clay is for and that's what I've used to
- 5:44scale my own business. So what Clay does
- 5:47is it finds the businesses and people
- 5:48that need what you sell and tells you
- 5:50how to reach them directly. So you
- 5:52describe the customer you want in plain
- 5:54English like find US companies with 10
- 5:56to 100 people hiring an AI automation
- 5:58engineer right now and get me the
- 6:00founder or head of operations at each
- 6:02one with their email. Clay will then
- 6:04build a list, finds the decision makers
- 6:06and researches each one so you know
- 6:08who's worth your time. A company that's
- 6:10hiring is telling you it's ready to buy,
- 6:12which makes your outreach feel smart,
- 6:13not spammy. And Clay's new go-to-market
- 6:15agent is about to launch and you can
- 6:17access it today directly through Clay's
- 6:19CLI in Claude. You tell it something
- 6:21like build me an outbound campaign that
- 6:23finds dental practices with five to 20
- 6:25employees and writes a personalized
- 6:27email to each office manager. Then the
- 6:29Clay CLI will build the whole thing for
- 6:31you. It builds the go-to-market
- 6:32workflows for you, the outbound
- 6:34campaigns, the scoring flows, the
- 6:36sourcing data, finding lists of leads.
- 6:38It can do it all, which is pretty wild.
- 6:39And the thing that powers all of this is
- 6:41that Clay connects to over 250 data
- 6:43providers. And by pairing that data from
- 6:45all these providers together, Clay has
- 6:47access to the best data in the world all
- 6:49in a single place. Now, on my channel, I
- 6:51talk a lot about problems you should and
- 6:53shouldn't solve yourself. And setting up
- 6:55250 data contracts is one that you
- 6:57should not solve. So Clay is free to
- 6:59start and no credit card is needed. And
- 7:01if you sign up through the link in the
- 7:02description, my community gets $250
- 7:05worth of free credits to get started,
- 7:07which you can get a lot done with. Some
- 7:08problems need one-time solutions and
- 7:10some problems are recurring. Finding new
- 7:12customers is a recurring problem and
- 7:14Clay provides a recurring solution. So
- 7:16that's about finding customers. The
- 7:17second thing that we're going to reverse
- 7:18engineer is about building tools, not
- 7:21apps. On this channel, I've stressed the
- 7:22importance of building for yourself, not
- 7:25for others. The reason is that if you
- 7:27build something that you will use, no
- 7:29matter what, you get value from it. And
- 7:31for me, this is actually where the whole
- 7:32buildpartner.ai roadmap comes from. I
- 7:34figure out what I need and I build it
- 7:36for me. And then it usually works for
- 7:38people who also want to get the most out
- 7:40of Claude. And so when you're building
- 7:41for yourself, the key here is also that
- 7:43you don't want to try and do everything
- 7:45at once. Every app or product that you
- 7:47see is really just a bundle of features
- 7:48and they're designed for the masses. So
- 7:50some features may be valuable to you,
- 7:52some features aren't. So when you're
- 7:53reverse engineering existing apps, I
- 7:56would avoid rebuilding entire things,
- 7:58but instead focus on rebuilding specific
- 8:01features or tools. And before you
- 8:02actually go ahead and rebuild anything,
- 8:04you have to make sure you understand why
- 8:06you're actually rebuilding it. It can be
- 8:08very tempting to fall into a dopamine
- 8:10building loop where it feels productive
- 8:12to recreate things, but in reality, it's
- 8:14not actually a needle mover. So if
- 8:15you're just rebuilding something to
- 8:17rebuild it, don't do it. There has to be
- 8:19a concrete enhancement that you want to
- 8:21do that it can't currently do, and we'll
- 8:23cover that a bit more in a sec. But once
- 8:25you've got a tool in mind, run it
- 8:27through our reverse engineering
- 8:29framework. First, pick one feature, not
- 8:31the entire app. Here is a prompt that
- 8:32can actually help you do this. So REA's
- 8:34own example does exactly this. It
- 8:36targets a feature, not the whole app.
- 8:38Their read me, their documentation says,
- 8:40understand how search works in the notes
- 8:42app. Show me the evidence and build a
- 8:43similar feature for my project. And if
- 8:45REA feels like too much, just use
- 8:47Claude. Second step, gather information
- 8:49about the feature. This is where I love
- 8:50either screenshots or a recording of
- 8:52using the feature. I've taken thousands
- 8:54of screenshots of features that I like
- 8:55to inform product decisions, and this is
- 8:57extremely effective. The third, design
- 8:59and focus on creating a minimum viable
- 9:01architecture. Depending on how you want
- 9:03to recreate the feature, you do not want
- 9:05to overcomplicate it. If it's one
- 9:07feature that stands on its own, have
- 9:09Claude make it as an artifact. Don't
- 9:10spin up a whole app and worry about
- 9:12hosting and all the other stuff. Just
- 9:14let Claude do it for you in its artifact
- 9:16templates. The simplest solution is the
- 9:18best solution. At this point, you should
- 9:20be ready for step four, which is rebuild
- 9:22the feature. Hand over the evidence and
- 9:24the design and let Claude rip or use REA
- 9:26to rebuild it. Five, iterate from there
- 9:28and make it yours. Back to understanding
- 9:30why you'd be recreating a feature. In
- 9:32this reverse engineering scenario, there
- 9:34has to be some enhancement you're
- 9:36adding, otherwise it's useless. So this
- 9:38step is probably the most important. Now
- 9:40building a tool is great, but the
- 9:41biggest time savings can come from
- 9:43something even simpler, and that's
- 9:45getting Claude to do a task for you. The
- 9:47third category to reverse engineer,
- 9:48automate tasks. And we're going to
- 9:50record the best version and then hand it
- 9:52off. The goal here is to reverse
- 9:53engineer how you complete a task. Then
- 9:56take that and systematize Claude to do
- 9:58it for you. Here we won't use REA in
- 10:00this step, but we're going to reuse the
- 10:01framework. And so the simplest way to do
- 10:03this is actually just ask Claude to
- 10:05review your past conversation history
- 10:07and flag any Claude skills you should
- 10:09create to systematize what you're doing.
- 10:11This will go through your conversation
- 10:12history and reverse engineer Claude
- 10:15skills based on your past activities.
- 10:17And if you run that, you'll have
- 10:18repeatable Claude skills that you can
- 10:20use for a task going forward. And you'll
- 10:21see in the prompt itself an optimization
- 10:23trick. When you do mine your chats, stop
- 10:25asking what did I use AI for, instead
- 10:28ask what did I have to send back. Every
- 10:30time you send something back, that's a
- 10:31step worth making into a skill, but if
- 10:33it got it right the first time, then you
- 10:34probably don't need a skill for it. But
- 10:36that is just level one of doing this.
- 10:38Level two is a lot more interesting. My
- 10:40favorite way to reverse engineer a task
- 10:42is to record myself doing it. Within
- 10:44Claude, you can record your screen and
- 10:46Claude will identify ways to automate
- 10:48what you're doing on the recording. But
- 10:50just clicking through the screens is
- 10:51only part of the process. You have to
- 10:53make sure to narrate while you work to
- 10:56give Claude additional insight into your
- 10:57thought process. That combines the
- 10:59visual component with the audio
- 11:01component. Claude will see what you do
- 11:02and then hear what you say. And as you
- 11:04do that, make sure to mention what to do
- 11:07and why. You want to explain it like
- 11:08you're teaching a human on an onboarding
- 11:10call. You would explain the logic behind
- 11:12your decisions. Then Claude will take
- 11:13all this data and it's trained to
- 11:15suggest programmatic ways to do the
- 11:16task. It won't just think about brute
- 11:18forcing it, it'll actually think about
- 11:20things like using APIs or MCP
- 11:22connections to automate this process.
- 11:24Then if you want to take it to the next
- 11:26level, you can record a rubric, not just
- 11:29the task itself. So you can record
- 11:30yourself reviewing the finished outputs,
- 11:32saying what's good, what's wrong, and
- 11:34why. And then Claude can take that to
- 11:36iterate and improve the skill over time.
- 11:38This is your way of teaching Claude the
- 11:40judgment, not just the clicks. So bring
- 11:42all of this through our framework. You
- 11:43want to first pick the task you do most,
- 11:45second gather the info through the
- 11:47recording process, third lean on Claude
- 11:49to design the automation or rebuild it
- 11:51and then grade it by running it
- 11:52yourself, and then five iterate until it
- 11:54matches how you do it and provide
- 11:56feedback directly. At this point, we
- 11:58focused on getting AI to automate tasks,
- 12:00build out specific features, and study
- 12:01successful businesses. But the biggest
- 12:03opportunity is to reverse engineer how
- 12:05industry experts think. Before we get to
- 12:07that, if this is your first video of
- 12:08mine, welcome the channel. But if it's
- 12:10your second or more, this is our
- 12:11anti-slap agreement. Everything here,
- 12:13the hours of research, the hours of
- 12:15editing, this is built entirely by
- 12:16humans for humans. So all that I ask as
- 12:19part of this agreement is you subscribe
- 12:20to help this content reach more people.
- 12:22And if you already subscribe, you're a
- 12:23legend, but make sure to turn on those
- 12:25post notifications. And as a thank you
- 12:27to everyone who watches, I give away a
- 12:28Claude Max subscription every video. So
- 12:30this video's winner is YTP Pepper for
- 12:33building an MMO RPG. For this video,
- 12:36comment below with what you're building
- 12:37to enter the next giveaway. Let's get to
- 12:39the next process we'll reverse engineer,
- 12:40which is strategic thinking. This is how
- 12:42you can reverse engineer how an expert
- 12:44thinks and works. Of everything in this
- 12:46video, this is the most important
- 12:47section. So I talked about earlier
- 12:49stealing like an artist, and this is the
- 12:51same idea at a more professional
- 12:52strategic level. And we're actually
- 12:54going to follow a Japanese martial arts
- 12:56learning practice called Shuhari. And
- 12:58here's how the practice breaks down.
- 13:00Shu, which means obey. Ha, which means
- 13:03break. And Ri, which means move away. To
- 13:05explain all three of these steps,
- 13:07there's a great example of this in the
- 13:09TV show The Bear. Richie spends a week
- 13:11working at a top fine dining restaurant
- 13:13where he starts by polishing forks.
- 13:14That's Shu. That's following someone
- 13:16else's system exactly, even when you
- 13:18don't understand why yet. Then he starts
- 13:20to get what makes that restaurant works,
- 13:21right? He sees how the pieces fit
- 13:23together, what matters, and how those
- 13:24practices could translate to a different
- 13:27environment. That's Ha. And eventually
- 13:28he brings the experience back to The
- 13:30Bear where he and the team build
- 13:32something of their own. That's Ri,
- 13:34taking everything you've learned and
- 13:35developing an approach that works for
- 13:37your own goals. First you obey, then you
- 13:39break, and finally, once you're ready,
- 13:41you move away. So how do you actually do
- 13:43this with AI? Well, first pick who you
- 13:45want to learn from. And this can be
- 13:46people in your own circle or people that
- 13:48you respect from afar. Then we want to
- 13:50gather. We're going to gather
- 13:51information from them so we can download
- 13:53their brain directly to AI. For people
- 13:55in your circle, just collect information
- 13:57you know about them already. And so you
- 13:58can use Claude connectors to pull in all
- 14:00their context, right? Their calls and
- 14:02conversations, their research frameworks
- 14:04and written work, their previous
- 14:05reports, their emails, scripts,
- 14:07presentations, anything else that shows
- 14:09how they think. And for people that you
- 14:10admire but you don't know, you'll need
- 14:12to work with public material. The
- 14:14simplest route is to use something like
- 14:15buildpartner.ai's expert advice, which
- 14:17is pre-trained on top professionals. But
- 14:19if the person you want isn't there, you
- 14:21can clone them with buildpartner's
- 14:22social tools, where you can run this
- 14:24prompt to search Instagram, Twitter,
- 14:25YouTube, Substack, and more to gather
- 14:28their material. And if you don't know,
- 14:29Build Partner is my tool, which is
- 14:31linked below if you do want to use it.
- 14:32Three, you then want to design a Claude
- 14:34skill using that context. The goal isn't
- 14:37to just give AI a bunch of documents.
- 14:39It's to capture how this person thinks,
- 14:40what framework they use, and how they
- 14:42approach decisions. For someone in your
- 14:44circle, you can build a skill around the
- 14:45material you've gathered. For someone
- 14:47who you admire from afar, you're
- 14:48building a model from their public body
- 14:50of work. Four, rebuild plus grade. This
- 14:53is where you really put it to work. You
- 14:54could ask, what would this person think
- 14:56about this situation? Give it a real
- 14:57problem to work through. See how it
- 14:59approaches the question, what it
- 15:00prioritizes, and what recommendations it
- 15:02makes. Then compare that what you know
- 15:04about the expert's actual thinking and
- 15:06work. Where does it get it right? Where
- 15:07does it miss something? What are the
- 15:09assumptions it's making? How can you
- 15:10actually improve this? And now this is
- 15:12where you start understanding the
- 15:13underlying method, not just collecting
- 15:15someone else's opinions. [music] And
- 15:17then finally, step five, iterate. Keep
- 15:19improving it as you find gaps, gather
- 15:21more context, correct assumptions, and
- 15:23refine the skill. Over time, you're
- 15:24moving through the same Shu Ha Ri
- 15:26progression. First, learning someone
- 15:27else's method, then understanding how to
- 15:29adapt it, and eventually developing your
- 15:31own. That's the real opportunity here.
- 15:33You're not just creating an AI version
- 15:34of someone else's expertise. You're
- 15:36using it as a way to build out your own.
- 15:38Now let's zoom out for a second and
- 15:39bring it all together. REA is a tool
- 15:41that rebuilds apps, but the real lesson
- 15:43is the five-step process behind it.
- 15:45Pick, gather, design, rebuild plus
- 15:47grade, and then iterate. It's the same
- 15:49loop every time, but you want to copy
- 15:51the process, not the property. And if
- 15:53you're still hesitant, Paul Graham put
- 15:55it very well. There's no better way to
- 15:56learn how something works than by trying
- 15:58to reproduce it. Nor does copying
- 16:00necessarily make your work unoriginal.
- 16:02Now I will say this tool REA is scary
- 16:04because it calls into question the value
- 16:06of products in our economy. So what do I
- 16:08think of the future of software? If
- 16:10anyone can rebuild how an app works in
- 16:12minutes, the code stops being the moat.
- 16:14So what is actually left that's hard to
- 16:15copy? Let's go back to what REA can't
- 16:18see. That's the back end, the
- 16:19relationships, the trust, and the
- 16:20judgment about what to build and why.
- 16:22When the cost of building a feature
- 16:24drops towards zero, the scarce thing
- 16:26isn't the software. It's about knowing
- 16:27what you need and having the taste to
- 16:28know when it's good. The winners won't
- 16:30be the people who build the most.
- 16:32They'll be the ones who know what's
- 16:33worth building for their specific
- 16:34customer. And building on that, a
- 16:36mindset I say to my team, we build
- 16:37things you can't just five code. So what
- 16:39are the things that make you special? So
- 16:41if it's Uber, it's not the app, it's the
- 16:43driver. If it's Airbnb, it's not the
- 16:45app, it's the hosts and the guests.
- 16:47There's plenty of things that make you
- 16:48valuable that AI just can't copy and
- 16:50paste. Now I touched on it earlier, but
- 16:52one problem you'll run into is this
- 16:54non-productive building dopamine loop
- 16:56where you continuously build things that
- 16:58have no real value. And the most
- 16:59effective AI builders know what to build
- 17:01and what not to build. And to get a
- 17:03better understanding of this, go check
- 17:04out this video where I share the five
- 17:06questions you can ask yourself to build
- 17:0810 times faster. If you're new here, my
- 17:10name's Austin Arkazy. Welcome to the
- 17:12channel and I'll see you in the next
- 17:13video. Peace.
About this transcript
This page contains the full transcript of Reverse Engineer Anything Does Way More Than You Think by Austin Marchese, generated from the public captions YouTube serves with the video. The transcript has 3,881 words across 569 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.