Por que criar seu próprio software pode virar uma armadilha

Author: Cortes do Rafael Quintanilha · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Por que criar seu próprio software pode virar uma armadilha · Channel: Cortes do Rafael Quintanilha · 10:11 · Watch on YouTube

Topics: Cybersecurity and software security, Software engineering with AI, Markets and economy

Summary

The speakers argue that AI-assisted reverse engineering is not entirely new, though stronger models and abundant access to tokens have made ambitious rebuilding projects more feasible. They caution that creating software is much easier than maintaining it, so companies should weigh potential savings against the effort of supporting custom tools. For durable products, technical advantages alone may be harder to sustain when competitors can use similar AI capabilities. They identify proprietary data, network effects, and especially distribution as important ways to stand out.

Key points

Reverse engineering existing systems to help agents rebuild them predates the latest wave of AI tools.
Improved models and plentiful frontier-model tokens make more ambitious software reconstruction possible.
Rebuilding software may save licensing costs, but ongoing maintenance can become a burden.
Businesses should prioritize their core product rather than spending excessive effort recreating tools that already work.
Open-source projects with active maintainers may be more appealing than individually built applications that users must support themselves.
Technical moats are harder to preserve when competitors can generate code using similar AI models.
The speakers name proprietary data, network effects, and physical components as potential sources of durable advantage.
Distribution and market presence matter: a technically strong product may fail if customers do not know it exists.

Chapters

Start	Chapter	Summary
00:01	Reverse engineering and what is actually new	The speakers compare recent AI reverse-engineering tools with earlier projects and discuss how better models have made rebuilding software easier.
02:10	The cost and practicality of rebuilding software	They discuss token availability and the possibility of replacing paid software, while noting that building a tool is different from maintaining it.
04:19	Maintenance, focus, and custom tools	Examples from enterprise software and common applications illustrate why teams may return to established products and prioritize their core business.
06:57	Technical moats and durable advantages	The conversation turns to why AI makes technical differentiation harder and considers proprietary data, network effects, and distribution as advantages.
09:07	Distribution and market presence	Using a Brazilian point-of-sale company as an example, the speakers conclude that even an equivalent or better product needs distribution to compete.

Quotes

building is one thing, maintaining it is a completely different story. — 04:19

If your business is a delivery SaaS, focus on building the delivery SaaS and take what already works for other areas. — 05:55

Proprietary data. — 08:35

if you don’t have distribution, it’s no use, right? — 09:38

if nobody knows it exists. — 10:11

Concepts and entities

Reversa, Sandec, Reverse Engineer in REA, MCP, GPT-4, Codex, OpenAI, Photoshop, Notion, Linear, Slack, Jira, Totvs, WhatsApp

Full transcript (original language)

00:01 This is a clip from the Limite Semanal podcast that happens every Monday at 7 PM with me and Vini from AI Coders. If you want to stay on top of everything relevant in the AI world, consider subscribing to the channel using the link in both the description and the pinned comment. >> This guy went viral, man, this repository, and everyone is talking about this thing, reverse engineering and so on, the future of software, and whatnot. Man, I’ll be honest, okay, Rafa? To me, there is nothing new here.

00:31 What’s new about it? They made a mechanism, an engine to make it easier for you to reverse engineer. Okay, fair enough, but let’s give credit where it’s due, man. Sandec here from Brazil already made “Reversa,” which is the same idea, man. Nothing to do with it, right? Reversa. Sandec. Look, Sandec made “Reversa” here, see? A strong repository too, 1,700 stars and all that. What is its role? And with a paper and everything, okay? The role of “Reversa” is for you to take an old system and create several specs for it

01:03 so you can take those specs and any agent can reconstruct it. So the idea of reverse engineering to develop a new system, man, isn’t that new. What came in the Reverse Engineer in REA here is that it has an MCP, it has an entire structure of, uh, visual capability and everything else to look at a system and remake it. Then new Photoshops started appearing. Did you get to see the Photoshop, uh, Open Photoshop and all that? >> Yeah, I, I saw some people talking a bit about that, right? >> Yeah, but like, I’ll be honest, Open Photoshop has existed for a long time.

01:35 Alternative Photoshop, uh, Open Office has been around for a long time too. So I don’t see this big revolution that people are talking about in the sense that this is just happening now. I see that people realized it’s possible to do this, and again, some people were already doing this, they were already reverse engineering. >> Yeah, I think people realized that you can >> take something that’s already finished and pass it to AI. Which is easier to do now with GPT-4. For example, I made my Age of Empires from scratch there using this. So it’s a trend, I think

02:10 you’re going to see people creating their own Photoshops and abandoning subscriptions. I don’t think so, actually. Do you know why, Rafa? Because the guy who uses Photoshop today doesn’t want to use an alternative version of Photoshop. The guy who uses Photoshop today is a professional. Because the guy who doesn’t use it, the amateur, he’s going to use an image generator, any random one, an image 2.5 from ChatGPT. He doesn’t need Photoshop. The guy who needs it is the one who goes into the little details of the work and adjusts it. That’s where I think it gets mixed

02:41 up along the way. What do you think, Rafa? >> Well, man, I completely agree with you in the sense that nothing here is new, right? Actually, this has been feasible for quite a long time, right? What I think might have changed are two things , okay? First, the models are getting better. So that allows them to be more ambitious in the task of doing the engineering. >> Exactly.

03:13 >> And the second point, which is something we also need to talk about, man, is the abundance of tokens. Because look, Vinis, honestly, after I started with APIs, with a plan, if I had my 200-dollar plan with OpenAI, I honestly don’t even know what I could do, because it’s so…I would have so many tokens available if you take Codex , right, OpenAI with its resets, man. So, you have an abundance of frontier model tokens today, starting from a month ago. It’s like, what am I going to do with all of this, right? >> Man. And a guy, a guy paying, let’s put it this way, for us here in Brazil, if

03:48 you pay for a 200-dollar OpenAI plan, that’s 1,000 reais, man. >> 1,000 reais per month. >> You have, you have tokens for a frontier model, okay? Like crazy, so, ah, 1,000 reais a month, Rafael, is expensive because we’re in Brazil, okay , fine, but we’re talking about you sometimes rebuilding software that you would pay 1,000 reais a month to use, you know? Or for you to rebuild software that you would pay a license for, you know, which would be, let’s say, Adobe products are expensive, for example. So,

04:19 >> perfect, perfect. >> So, there’s a lot of stuff you could do , right? There’s an interesting point, Rafa. Since I’ve moved into this space of rebuilding software instead of using off-the-shelf solutions, I’ve done everything from CRMs and basic systems to marketing automation and more. Man, you can do it, but building is one thing, maintaining it is a completely different story. And what are we going to start seeing now? A bunch of people creating their own software; we already see that in companies, right Rafa? Usually, the market starts at the Enterprise level, and then it trickles

04:50 down here. >> I’ve seen several Enterprise-level companies that rebuilt systems internally, created their own, and some >> because it seems like—sorry to interrupt—but Enterprise guys usually spend a ton of money on software, right ? It’s a cut where the guy will say: “Oh, I’ll save 30,000 a month on the bill, depending on the tool.” >> But there are companies where, in some steps, it made perfect sense and they actually ditched a lot of tools. But there were other cases where the team said: “Man, this is giving us such a headache to maintain. It’s more worth it to go back to Notion, for example, or where?” Or to Linear, for instance,

05:22 right? So, if you stop to think about it, now we’re talking about binaries, right? Binaries that REA has in this proposal, but simpler systems, so to speak. Take Slack, for example, man, you could have built a Slack with AI a long time ago. So much so that they did . Did any solution actually take off? >> It hasn’t really caught on yet, man. Take task management, Linear, Jira. Why is there still a bunch of people using Linear, for example? Because it’s a good solution. And that’s where it gets complicated when it comes time to maintain. Because my focus today, and something I always ask, is: What is

05:55 your product? What is your business? “ Oh, my business is that I’m developing a delivery SaaS.” Cool. If your business is a delivery SaaS, focus on building the delivery SaaS and take what already works for other areas. And then, once it’s up and running there, then you can cut costs here. Now, if you’re focusing on a Delivery SaaS but spending a huge amount of energy building your own harness, man, I don’t know if that’s the smartest strategy right now. So I think what we’re going to see happen is a lot of that, Rafa. we’re going to have people, uh, excited, because it’s really cool to

06:25 be able to reverse engineer stuff and generate your own application. >> Uh, but which ones will actually be used? I think we will see more open source projects appearing and then yes, there will be a strong base maintaining them, but there won’t be so many, so many tools, because at the end of the day people will look and say: “It’s cool that I built mine, but I don’t want to maintain this thing.” Let me find one that is being maintained, that is getting bugs fixed and is evolving while I focus on my own thing. That’s how I have >> for sure, >> seen it. >> Nice. Yeah, I just wanted to extend

06:57 this here to say one thing, okay? Then you’ll see, right, people saying: “Oh, software is dead and all that, uh, the software industry in general.” Uh, and although it’s true that you can reconstruct things, as you well said, there are some issues that I think people need to be paying more and more attention to, Tavin? I want to bring this up here, oh, uh, let me put it here on my, on my screen, >> look. I’m the one on the screen here, right? Thanks. >> I wanted to bring up here, oh, this post from this, from this guy here, Nikita Bia, he was for a long time, uh,

07:30 I don’t know if he’s a PM, right, but he worked there at the, at the ex, right? >> And then he said, oh, and technical moats, right? So it’s that, it’s that thing that only you have, right? That which protects you from the competition , right? So he said technical moats, so basically they’re gone nowadays. So, it’s very difficult nowadays for you to have an advantage in software that is, if you look, you have your secret, right? You have the secret sauce that only you have and nobody else has the

08:01 technical capacity to do what you did. That’s very difficult. >> That was a long time ago, right? >> It’s long gone, man. It’s gone. For example, I remember with search, its index, which was like, the best of all. Uh, you know what’s very good too is WhatsApp, man, the WhatsApp experience, the WhatsApp messaging is, is, is an app that is extremely hard to operate at its scale. It’s an app, right, with absurd reliability, right? So, the time when you could just say, “hey, technically, I’m superior,” is gone.

08:35 Why is that no longer true? Because basically everyone does things with us now, so everyone is capable of generating code from the same source of inference, right? So if we can’t stand out technically, what do we have to use to stand out if we want to build something meant to be durable? I mean, we’re not talking about internal apps here, we’re not talking about “vibe coding,” we’re not talking about throwaway code. Disposable code for an app I create to solve a problem and then throw away.

09:07 >> How do you solve this? Proprietary data . Perfect. >> Network effect or a physical component, right? So if we set aside the part, >> Dig a little deeper, Rafa, I’d say a bit more about distribution. >> Yeah, that would be the network effect, right? Network effect, distribution. >> Yeah, right? But I’m talking about distribution. Imagine the big companies today, for example, here in Brazil, you have Totvs, which is a giant in the retail and POS sector. Man, to take Totvs out of the market today, you’re not going to just code a POS and take all of Totvs ’customers. The problem is

09:38 , if you think about it, there are probably 50 million POS systems better than Totvs’, you know? The point, >> The point was never that, right? Exactly. Exactly. So, I think in that aspect, the guy who dominates the market today, who’s there, has a strong distribution, has a strong presence, which is what we’re talking about, it really is the network effect, but going beyond distribution and all that, man, that’s what rules the market today. You can have the same solution, identical, if you don’t have distribution, it’s no use, right? So it’s not enough to just have a beautiful product, well-made

10:11 technically, if nobody knows it exists.