Building a Machine | BLG Thinking #3
Third episode of my SaaS journey. Rebuilding the entire thing in 2 weeks and more...
Ladies and gentlemen, welcome to this third episode of my BLG Thinking podcast. Now you can listen to the episode podcast by clicking on the audio player directly into this post.
TL;DR
This summary has been generated by AI based on the podcast transcript.
Create: Platform Migration & Infrastructure Overhaul
Complete migration from Vercel to Cloudflare in 2+ weeks
Rebuilt entire SaaS infrastructure from scratch
Performance issues: workflows taking 10-15 minutes, cold start delays of 30s-1min on Vercel
Cost concerns with Vercel pricing model
Cloudflare advantages
Faster performance, no cold start issues
Significantly cheaper than Vercel
V8 isolate limitations bypassed with new Cloudflare containers
Built comprehensive platform (heyblg.com)
Substack/Shopify model: single platform, multiple specialized AI stores
Automated infrastructure for creating new AI products
Focus shifts to workflows only, platform handles everything else
New Product Launch: Andy SEO
Launched andyseo.com - specialized AI for SEO workflows
Core philosophy: “AI + deterministic workflows + right data sources”
Five main workflows/products:
Brand Strategy
Website analysis, competitor research, voice definition
6-tab strategic document output
Content Mapping
Topic pillars based on brand strategy
Keyword Research
Domain authority assessment, relevance filtering
Live Google data integration
Content Classification (4 types):
Transactional (write it)
Defensible (strong opinion/original data)
AI summary optimized
Non-defensible (avoid - AI replaceable)
Content Brief
Article Generation
Batch processing
Results validation: Generated 90-100 articles for Andy itself, 30-40 for Isabella
Each article includes tone, internal linking, technical SEO elements
2-3 hours total time investment per batch
Explore: Strategic Vision: Building Machines
Leverage concept: same input, exponentially more output
Three historical leverage methods:
Innovation/Technology (wheel → code/AI)
Human labor (employees)
Capital (money/debt)
Volume requirement for scale across all leverage types
Focus shift: building machines that build machines
Andy creates content volume for other products
Distribution-focused products only (pausing Enzo development)
Volume of customers > product quality for business success
Level up: Next Product Pipeline
Three new distribution-focused products (2-3 week timeline each):
Social Media Content Creation
Audience/niche analysis, hook/CTA optimization
Human creativity + AI execution/testing
Target: 10 variations per day
Video/Content Editing & Automation
Multiple format variations from single idea
Automated posting capabilities
E-commerce Video Ads
AI-generated videos using new models
Testing phase → double-down strategy
Apply to existing 10k/month e-commerce business
SEO foundation established but long-term game
Need faster volume generation for immediate results
Social media organic content identified as fastest path to volume
Transcript of the podcast
Hi, All right, ladies and gentlemen, welcome to BLG Thinking number three. Well, this episode is a bit late. Because we’ve — well, I’ve been doing quite a lot of things recently, so we are going to talk about this.
1. Create: From Vercel to Cloudflare
Let’s go as always with the first section which is the Create. Well, it’s been a little bit more than a month from the last episode. And, boy, there has been a lot of work. The big change was that I completely migrated from Vercel to Cloudflare, the entire platform. And actually, what I did was I rebuilt from the ground up, from scratch, the entire thing and my entire infrastructure for my SaaS and for my AI SaaS.
So that was quite a lot of work, but at the same time, it only took me, like, two weeks, a little bit more than two weeks to rebuild everything from scratch. So given that into account, that’s not too bad. I’d say so I obviously use Claude Code to do all of this, and, man, that was — that is pretty impressive.
Why migrate — performance and cost
So let’s go just quickly in the technical details. The first reason I wanted to do that was for the performance issue and the cost issue. So, basically, even though it was just at the very beginning, already I kinda saw two problems that I was going to have very quickly.
The first one is the price issue. I always knew going to Vercel that was going to be an issue. I didn’t expect it to be that soon. However, the big problem was really the performance issue.
The workflow concept
So to talk about this, my entire infrastructure is relying on one thing, which is the concept of workflow. So what does that mean?
That means, from the UI perspective and from the actual UX perspective, it is just an AI chat as if you were classic, something you can find pretty much everywhere nowadays in any kind of AI chat. The big difference, and this is where the whole vision, the product vision is going into place, is that I have what I call products inside, which are actually workflows.
So basically, there’s an AI chat where you can talk and just make some very easy tools like read and write and whatever. But the real deal is the ability to use that I have to create specific workflows. Let’s say I have a workflow that — this is for example for Isabella, my research worker — that can batch process multiple YouTube videos. And under the hood, what is happening is that:
I go and take the YouTube URL link that the user gave me
I then fetch in parallel all those videos
Get the transcript
Then make some AI calls to summarize the transcript and do whatever
And then create the output results
And this is just a very simple and easy workflow. Another one I have — by the way, this is something we’re gonna talk about, the new product, because there is a new product — is about making API calls to get some specific data. So then we are going to make first an AI LLM call to classify the data, and then a second one to output and show the result to the user.
But as you can see, a workflow basically contains multiple steps, and all that is hidden from the user, which is the value of the product. I basically design specific workflows that the user can use, and it feels like magic because you’re just talking to the AI at this point.
The technical bottleneck
The big technical issue here is that the workflow can take minutes. Like, for example, the new workflow that I have, which is about SEO — by the way, let’s go directly into that. We — I will talk about this a little bit later in a few minutes.
But, basically, this workflow can run for ten minutes if not fifteen. And this was going to be an issue from the start because Vercel doesn’t allow that. And so I tried multiple ways. Specifically, there’s a product that is called trigger.dev that allows me to do that to some extent. That was good, but there was a lot of issues in both cases, and there was a big problem which was the pricing — at some point it was going to cost me a lot of money.
So this is the first problem. The other option was to go the old way, so just to have a server and go to Render.com for example. That might have been an option but I don’t think for the long term that would have been much better, and that would have caused a lot of other issues as well.
Discovering Cloudflare
So I decided to go to Cloudflare. To be honest, I didn’t really know all the different services they were offering. I just knew from, you know, DDoS point of view, basically the CDN that they were offering, and they are known for that.
But, actually, it is genuinely very good, impressive, performant and cheap. The only downside is the V8 isolate problem, which is a performance issue, because this is a primitive of Cloudflare. But now with the new Cloudflare containers, you can easily bypass that and that is working pretty well. This is pretty much impressive.
So this was the big reason why I went. And this should have started just as a one day audit to see if that was worth it. Ended up into a two weeks complete recreation from the entire thing. But that was impressive. And now everything’s working out of Cloudflare. It is not only very reliable and working — from the performance, this is way faster.
Cold starts went from 30-60 seconds on Vercel to basically zero on Cloudflare.
Yeah, that was also — there was also a speed problem. So another problem that I had with Vercel is every time I was launching a workflow it might take thirty seconds to one minute to just launch it. So I had a real cold start issue that I do not have at all with Cloudflare, and this is very impressive. All of that with keeping everything serverless. At this age — impressive. That is just impressive.
So, anyway, without going too much into audio — too much technical — but that was very interesting because I had to learn every primitive of Cloudflare, how everything is working, how everything is connected to each other. And so on and so forth. So that was a big thing, and now it is working, man, just fine.
Building the platform — the BLG vision
And this is even better because I used that time to not only migrate the same thing, I used it to actually build a platform, an entire platform, and this is basically my platform vision.
So my vision about the product is that I’m building an infrastructure and that is the platform. So this is the entire UI and UX, and the big infrastructure about the workflows and stuff like that. This is one thing. But then the key here is that I can basically create any kind of specialized AI in pretty much any field. And the only thing left to do is to write and create the workflows themselves, which are the value.
But the entire things outside of the workflow — so the AI, the chat interface, the workspaces, the account, the settings pages, the even the drivable, which is a new concept — everything else is the same.
I’m creating specialized AI chats for very different use cases. This is really about giving AI a superpower through specific workflows.
So that is the whole vision of the product. And to do that, I built an entire platform which is way easier for me because this entire platform is working just like Substack. Basically, this is based on a mix of Substack and Shopify model where:
Substack is the platform and then I can now create a new “store” where I put my content inside of it
Shopify is the platform, but I can create a new store where I have my ecommerce store and I put my product inside of it
But the underlying tech is the same. This is pretty much the same thing. I have a platform which is, by the way, called heyblg.com. That domain is not going to be used a lot because you’re not going to be connected, at least not right now. This is just for me. And maybe one day, will open up to different people, but whatever.
So this is the actual platform. And then I can basically create the content itself directly and just focus on the workflows, and the rest is automatically done. That was the whole idea and Cloudflare is not only helping me do that — it’s actually very easy to do that with Cloudflare because there’s a lot of things that I can automate.
Introducing Andy — the SEO specialist
And on top of that, to prove my point and the whole point of the platform — to be able to create new specialized AI in different fields — the new one is the SEO product that I called Andy. The URL, it’s already online: andyseo.com. And here, this AI is specialized well in SEO. And Andy allows you to basically be your head of SEO and just do everything.
And that is very much the key here. The whole idea of workflows is that I can do a lot of stuff. This is not just asking AI to do something. It is mixing AI — and this is, by the way, the whole vision of the landing page, this is what it says:
AI is a technological shift. They say it will automate the work. Truth is you spend most of your time teaching the AI how to do the job. The model isn’t the bottleneck anymore. What’s missing is a deterministic workflow and access to the right data sources.
And this is exactly my vision. This is exactly what I’m going to do and sell.
Andy’s five workflows
So if we go back to Andy and to SEO, the workflows there are five workflows, which I call products.
1. Brand strategy
The first one is brand strategy. This workflow is going to ask you your website and some questions:
The website URL, what you sell, the country you want to rank in
Then the workflow will fetch all the main pages from the home page (product pages, pricing pages, whatever) and read all of that
Then it will ask you three to eight questions focused on your SEO goals, your audiences, your target audience, your expertise, your hard opinions, and what you refuse to write about for SEO
Then it will ask you for your voice — paste a few examples of your own writing
Then automatically, based on this data, it will run real competitor search — running Google searches for the topics that you wanna own, and see who are your SEO competitors
Then it will list the pages it read about the website for the internal linking
And based on all of this, it will write up a strategic document that understands your brand, your products, your audience, your tone, your expertise, your internal linking, your competitors.
So that is the first workflow or the first product. And that will be the base of everything. And as you can see, this is very step by step. This is very deterministic. There’s steps.
The real value of AI is not just about putting AI everywhere — it’s about mixing what we always knew about tech, deterministic workflows, with an intelligence layer on top.
So the output here is that you will have a very big document with six different tabs. That is going to be the base of everything else because here you really understand your brand, your audience, your voice, your opinion, your competitors, your site structure, and everything else.
2. Content map
Now the question is, what is the point of doing all this? The point of doing all this is to follow the step by step best practices to write SEO. The same way if you hire a SEO agency, the very first step is not about searching for your keywords. This is about understanding who you are. What is your brand? What is your product? Who you are? What do you believe in? What is your tone of voice and stuff like that?
Then we will create some content maps and different pillars that you wanna own on the topic. And there’s also a workflow for that.
3. Keyword research
Then we have the keyword research. So once you have done the brand strategy and the content map, you can run another workflow which is again very small, that will first of all:
Read your brand strategy and know who you are (very important)
Pick the scope and the topic you wanna talk about and write about
Then the AI will be smart enough to understand your domain authority — are you a new domain and a new brand in the space? Are you already established? — to understand which keywords to focus and to rank for
Because the problem is — and this is very important in SEO — it is very technical and strategic at the same time. A lot of people just go to search for keywords and after that, they would try to write an article on this. But you have to understand if this is related and relevant for your brand and based on your authority.
4. Wide discovery
Step four, wide discovery. We’ll actually do keyword research by looking at real data — not just doing an LLM call at this point. We’re looking at:
Live Google results
People also ask
Related searches
Conversational queries
Then we’re gonna filter that. And here we’re using AI to filter the research to drop anything that doesn’t fit your audience and that doesn’t make any sense. Then we check the volume, the difficulty of the keywords, we check live data.
And finally, very important, we are going to define whether or not the keyword and the article is worth writing. Very important, especially in the AI SEO era, where, in my opinion, there are four different types of content:
Transactional — and you should write it. These are people that wanna buy your product. Definitely write it.
Defensible — this is the content you are going to win because you have a strong opinion, and AI will never be able to create and generate a strong opinion. So here you can win, on depth, opinion, on original data.
Defensible with AI summary — this is a bit different. You wanna write a topic, but not for human. You wanna write this topic for your content to be quoted inside the AI summary, instead of clicked from below it. Very important.
Non-defensible — maybe that subject will be easily replaced with a generic AI, that AI can write itself.
A big part of Andy is that Andy is smart enough to tell you what to write and what to not write.
5. Brief and article writing
So this is just like if you hired a SEO agency, except here, this is the AI that is doing all the work. From understanding your brand, researching the keyword, doing the content brief for an article, and finally writing the article. And you also have the batch mode that can generate the briefs and the articles by themselves.
The proof — using Andy on itself
And, well, again, from the technical point of view, it only took me almost two weeks with the marketing website and stuff like that, but it only took me one week to focus on the actual workflows. And the workflows are the core idea of what I’m selling. And all the rest is already done by the platform. And that is going to be impressive because now I’m going to be able to create multiple specialized AIs very fast.
And finally, just to end up this Create part, I’ve been able to use Andy for my own product because I’ve used Andy on Andy itself to create the blog and it generated me about 90 to 100 articles. And this is very important, each has my tone of voice and understood the actual brand. So in this case, that was Andy itself, which was kind of funny.
It searched the right content, the right JSON MD from the technical point of view. It wrote the article with the H1, H2, whatever, the internal linking, the AEO, the pillar contents, everything else. So that is very impressive that I’ve been able to automate the entire thing while keeping the quality and the value very high. And, again, it took me about a few hours to do that.
And you can even go further if you have your own documents — you can give everything to Andy and it will use that to really create content from your expertise, and that is going to be a very big difference.
I also used the same for hey Isabella. So I just used Andy to create blog articles for Isabella on a very different subject. And turns out it was working just fine. It created about 30 to 40 articles, each based on a very specific keyword, based on the strategy, based on different pillars and internal linking. Boom. Everything in about two to three hours to do all of that.
So that is the new product.
2. Explore: Building a machine
Now we are going to move on to the second part. We are going to talk about building a machine. Well, and this is really going to be my focus because I guess this is every engineer’s dream — to build a machine that is working for you 24/7.
And there’s a lot to talk about here because this is really my view and my vision with AI: that you can delegate a lot of things with AI without sacrificing the quality of content. That is going to be very much the key and very much important.
AI can replace the execution part of the work. However, for creativity and creation, the human is where it just makes so much more sense.
And this is why, if we go back to Andy, a big part of the quality of the content is you. What do you believe? What is your opinion? What is your product? How do you write? Because if you ask the AI to write instead of you, that is not going to work. And that is going to create generic results. That is why the focus on Andy is to first understand you — the input is still you — but then to leverage and scale that.
The three forms of leverage
So let’s go back to the idea of building a machine. What does that mean? I think it all comes down to the idea of leverage, which means to use the same amount of force but to have way more power.
And if we think about this in the past, in the history, this has been done — and which is still done — by three different ways: innovation, human, and capital.
1. Innovation / Tech
So what does that mean? Let’s go back to leverage. So it’s having basically more output with the same input. How do you do this? You do this by leveraging something. In my opinion, the best example is the wheel. The wheel is to some extent tech. It’s a technology. It’s an innovation. That allows you with the same amount of effort to do way more output than without it.
So that was one way of using leverage. And we see this nowadays with tech and innovation. However, today, it is by code and AI and online. There are different ways, but this is mainly what we’re gonna talk about.
2. Human labor
The second way is obviously human. So this has been done in the past with slaves. But if you think about this, is pretty much the same thing nowadays with employees, except they may have a little bit more rights. To the point that you are using other people’s labor to get something out of production. If you think about a company, this is pretty much the same thing. You have a bunch of people working for you and you have, with the same input of yourself, more output thanks to other people’s labor.
3. Capital
And the third way is obviously the capital. The money. Well, if you already have the money yourself or if you can borrow money and be in debt, well, you can use that to leverage and have more output than you would do alone.
Scale, volume, and the machine
So these are the three ways. And that combines into the idea of building a machine that will work for you instead of you working directly.
And if we think about this a little bit more, that is the idea of scale, and scale is the idea of volume. Like, you need some volume directly or indirectly.
So what does that mean? That means you need to accumulate a volume of resources in order to scale, in order to have leverage, in order to build something:
Real estate — you need to have some volume of money to access the real estate and then maybe do rental and stuff like that
Product and entrepreneurship — you need some volume of clients to really have something working for you
It all comes down to the idea of building a machine for you, and the big difference is all about the volume.
Building machines to build machines
And if we go back to this, for my vision and what I wanna do here, I want to build machines for myself. That work for me, that basically make money for me. And I will do this by building machines to build machines.
That does make a lot of sense, but this is basically what I’ve been doing with Andy. I’ve built a machine that is able to create volume. So in this case, this is SEO articles for my other machines. Do you see the logic?
And now — and this is going to be the big change here — I’m going to stop focusing, at least right now, on products that are not directly linked to distribution. So I had another product, hey enzo, which is still there and still working, but I’m not going to focus on it too much at least right now because the priority is not based on distribution.
Distribution is everything
And the goal here, if you’re thinking about volume — what is the difference between a $100 a year business and a $1,000,000 a year business? It is not that much the product. You may have the best product in the world — if you do not have customers, it doesn’t work.
The big difference is the volume of customers you have. It is all about the distribution. Always.
We always come to that conclusion. So this is why — and this is, again, all about volume, all about scale, all about leverage, all about building a machine. So now, I really need to focus on using my leverage, using my infrastructure, using my platform and my products to build machines to build machines.
So I’m going to focus on creating new products that will automate the distribution work. And if I can do this, that would basically mean I’ve achieved distribution, and I can do this for pretty much any kind of business. That is going to be the new focus.
3. Level Up: automating distribution
Which makes the transition for the third and final step, Level Up. What is going to be the new thing that I wanna do? Well, as always, as I’ve said before, my new focus is going to be on building new products with the idea of using those products, using AI, using automation, code, using social media, and using algorithms to build a machine that will build distribution for me.
So that was the first step with Andy and SEO. I think SEO is a very good start. It is a foundation. That allows you to have a clean foundation for your website. However, especially SEO, this is not magic. You cannot go all crazy on SEO and just rely on SEO, especially in the beginning, if you have no domain authority. So you can do backlinks. You can do that kind of things. But even if you do that, that will be very long. That will take a lot of time. This is not something you can go fast on.
So I think you can scale very good with SEO, but SEO is really about the authority and the older you are — at least the longer you exist in the eyes of Google — the better you are. So this is very a long term game.
Three new products to crack distribution
This is why next, I’m thinking about doing three new products to really focus on creating volume. These are going to be about content creation, so social media organic content, which I think is going to be the key here because you can really go fast. You can achieve an amount of volume very fast.
I need to really learn more about this. But I think about outliers, about understanding the hooks, the CTAs, and obviously about two phases when you’re doing this:
Testing phase — when you have to test multiple formats
Double down phase — when something just works
But as always, this is all about volume. This is all about testing a lot of different things. This is all about consistency as well to double down.
Product 1 — Content ideation
Well, what if you can automate a lot of that work? What if you can create a content product, a machine, that will allow you to analyze your niche, your audience, and what is the kind of content they consume?
And then this is very important. This is where the machine and the AI is very good. However, when it comes to having the idea — this is where the human has to be very good. The AI might give you some ideas, but it will never be able to be creative. This is where the human has to be, and you have to combine the different hook formats and whatever with your unique idea. This is how you are going to be able to differentiate yourself.
Product 2 — Editing automation
Then the second product is all about editing. Because this is basically just like ads — if you’re doing ads, you know that the content might be the same, but you wanna do a lot of different variations on different hooks. And what if you can automate all of that? Well, I think we can do that. Because the idea is human, but the actual edits can be automated.
That would be the second product. And, obviously, then inside that product, we want to do something that can post automatically for us.
And, again, who do you think is gonna win? Someone that is able to have a workflow and a machine that generates 10 a day.
The key here is: can I keep the quantity high while keeping the creativity human?
Very, very important. Because again, the creativity should not be from the AI. Absolutely not. The creativity should be from the human.
So my dream here — and I think this can be done in one or two weeks — I wanna have an AI that when I say, “Okay, here’s my idea,” that AI will analyze all the different outliers and content of my specific audience and give me more ideas. So this is the brainstorming step. Then I would say, “Okay, this is my script. This is what I wanna test. And this is the core value of my content.” Then I film it myself.
So there’s a lot of parts that are still very human and manual, and that’s fine. This is about optimizing what the AI and tech is good at and what the human is good at:
Human — being creative
AI and tech — automating and analyzing a large amount of data and volume
Once you have that script, that idea — well, you can test a lot of different things. Again, this is all about marketing. It is audience. What is your audience? What is your message? So that is very much linked to human and creativity. But once you have your message, you can test a lot of different formats of hooks, of CTAs, and just test. And this is volume. At the end of the day, this is volume, and that can be automated.
But again, I have to be smart here to create something that does not reduce the quality of content. And this is why the creativity has to be human.
Product 3 — Ecommerce ads
Okay, those are the two new products I wanna create. The third one, which is going to be a little bit similar, but this time is going to be for ecommerce, for ads.
Again, remember I have an ecommerce business on the side that is making almost 10k a month. And we need to double that down. And, again, this is very important — this is what I understood about this.
What is the difference between two years ago when this business was making 2k a month and now it’s making 10k a month? The product has not changed. Okay, the website has changed. So it’s a little bit better. What has changed as well is the volume. Now we have way more people that discover us through SEO. It’s all about distribution, volume and distribution.
This is why I’m so obsessed about distribution because I know this is the key. I just need to crack it. And when it comes down to ecommerce, what I want to do is go a little bit more into AI generated videos. There are new models that have been coming out lately, and if I combine those new models with — as I’ve said — the logic and the workflows of understanding your audience, understanding the message, and just doing a lot of testing here to test what works and what doesn’t and then double down on what works.
And what if you can do that at scale? Just automate the entire thing, and again, building a machine that will be able to build machines. That is going to be the next focus for the next few weeks. I think I would be able to do that in two to three weeks. So let’s see how that turns out.
And if I can do this and if I can crack distribution, well, pretty much everything else is going to go up from there because, as I said, the product is there. The platform is there, the infrastructure is there. I really do believe there is a lot of value into the product. But again, product is one thing — the most important thing is distribution. So let’s go figure that out.
Mood
And for the final word of that — what is the music of the day? Well, that’s been Logic — “Intro”. I’ve been a big fan of Logic and NF, you know, it’s been a few years now. And this song, I didn’t know — I just discovered it a few weeks ago. This is very good and that is really much my type of song. This is really much what I like to listen to.
And I think there’s two kinds of people:
Those who listen to songs and music based on the vibes
Those who listen based on the lyrics
I am definitely on the second type of people.
So, anyway, this is it for BLG Thinking, episode number three. I hope you have a good day, and we’ll see you later.