Everyone Codes Now: What Happens When Marketers Get an IDE — Vladislav Sopov | GetBlock

Thank you. Thank you so much. Thank you guys for coming. Thank you. Hey everyone.

Uh good to be back. This is my actually second time at Ethereum Bilgrat and I have been looking forward to it since the last one. I run marketing at Gatblog and for the next 20 minutes I want to talk about something this room is a perfect audience for. Uh quick on who is talking. Uh my name is Vlad nine years in crypto seen from three seats uh marketer which is my job now uh before that a journalist from you today to be encrypted to forbes and for a while an independent analyst.

So uh simple question to start uh everybody is doing AI in marketing. Everybody's got a milestone. Everybody has got a drought. Uh why sit here for one more pitch? And here is my answer.

Uh almost every AI story falls into one of two buckets. You learn one from Techrunch, you learn the second from Wall Street Journal. Both are great read and neither one helps you. Let me show you market one. Uh the tiny startup that automated all of its marketing and it doesn't need marketers any longer.

You have seen the post. Uh the problem these are mostly revenue team. No real users, no product uh market fit, nobody showing up. They went from one click to 10 and break. You can't tell if it repeats because it has nothing to repeat yet.

Uh, bucket two, the opposite, the giants. They tell us about the marketing director straight from OpenAI 30 grand a month uh per employee on tokens and so on. It's impressive, but it's more frustrating than useful uh because you can't copy a balance sheet. Nobody nobody from us here is standing to start a in-house ML lab next quarter. So where is with everybody else?

The team like the most team on this room and in this conference uh who have thousands of paying customers and grown who have confirmed product market feed uh who have real brand with real paying users nobody write this case study. It is not sexy enough for tech blocks and not big enough for big adult business press. This gap is why I am here. Where are we going? Three things I want to tell you.

Just three stories. Uh how to get teammates who have never touched code building real paging services. How to get the language models to know your product. And how to measure whether either one works. My promise you can replicate all of it in about a week.

Not a quarter but just a week. Go build one and I will be happy. Part one. Uh what happens when you hand a marketer a real uh development environment. This goes three ways.

This can go three ways. It can be useful. It can be waste of time and it can be dangerous uh for your product. Uh the setup is the decision. Why want this?

Uh three reasons and they stack. Uh with marketers coding, you ship faster. You test hypothesis faster. And the most important one, you stop taxing your engineers with marketing leg work. Every hour they spend moving pixels is an hour stolen from the things they only they can do.

Uh the actual setup u from gblock. Uh you just need to install the coding agent. I suppose you already done this for your laptops cloth code on codex and so on. Uh then open it ask it in plain English to install everything else. G nodes note version manager and so on.

Nobody in my team uh learned what package manager is. But the only human skill here is the opportunity to ask. Now access real access comes with real guard rails. Everybody gets a GitLab account on their work email joined to the get block organization through single sign on. A secure key is generated on the laptop and registered with our GitLab.

Uh we have written guides for Mac and for Windows. Uh but guys just don't email a marketer the steps and wish them luck. uh your CTO or someone from develop should sit down next to you uh your marketer with at least online and uh put it together. Uh 20 minutes side by side and we are good to go. Last piece from empty folder to working codebase.

Uh you your prompt engineer swipe coders open the agent ask it to clone a main repository to commands and environment build itself. It's done. Now you're thinking this is where this goes wrong. A marketer pushing to real life production and this is where review saved you. Every change at getlo gets a code review through harmes or open claw set up by our studio Nikita and it talks to us through slack.

You can see here like our deol lime asking in slack to approve her merge request and it says it good comments fix approvals all in slack. It works in 99% of cases. So what do engineers actually do in setup? They on board people, they teach and they own what I call the full prevention layer. They stop doing the small stuff by hand and design the guardrails the system that lets everyone else do it safely.

What comes out uh what we can build with it? Landing page, lead magnets, small services and so on. Uh you can see our RPC speed tester built completely by marketer by our colleague Natalie. [snorts] But my favorite is a quick one is a quite one fixes. Uh any marketer can now patch a typo or a broken link on the website and that's doubly huge.

If your site runs on something custom or homegrown, some CAS your engineers build themselves. The bottleneck disappears. Uh sorry uh guys quick got check maybe show me your hands. Uh have you already engaged your non- tech people non-engineers pushing to live sites or would you Yep. That's great.

That's great. And who is a better builder marketer or product manager? Non tech product manager. [laughter] Why? Why?

Okay. So,

yeah, of course the answer is PM. So, uh part two guys, that's that's my favorite part. And I'll be honest, this is my Roman Empire. I ran first row is pugments back in mid 2024 and I still can't put it down. I called it LM visibility but then I found a fancier word and here goes LLM discoverability.

So now I sound smarter at town halls. Two things do the models know you exist and when they answer are they sending you a customer? One thing I'm leaving out today is paid uh advertising and open a ads. Uh I haven't tried it yet. Uh sorry.

Uh the good news I start from the good news. GEO is the CEO. It is almost annoying how simple it is. Uh getting found by language models is mostly the same search optimization we have loved and hated for years. Clean site, no broken links, no dead pages, easy for crowers to read.

So my advice sounds basic. Uh run a real audit. Uh fix what it finds. But two things matter for me. Two things matter.

Block and docks. Block and dogs. I repeat twice because those are your cheapest tickets into the models. Everybody killed their blogs because humans stopped reading them. But the main reader now is machine and it reads everything.

So far it's old news. None of that is new. It all worked for Google for a decade. But let me show you two things that are actually different. Our two real new campaigns.

First one get block has genuinely great numbers in Asia. Our Singapore C cluster was the fastest Salana connection in the region and there is independent benchmark of mine. uh 36% faster than Alibaba cloud. I can say it was fantastic but it was completely useless because the models had no idea if this was true. So I ran a bet uh my theory the models are hungry for numbers non-adjectives.

I wrote dense almost robotic articles about latency city by city about uh from Hyderabad to Tokyo. Uh real request samples uh explain how transactions land on Salana and I placed them across uh six or seven uh second tier Asian media outlets about 20500 grand. The editors were annoyed. Uh they said it reads like a manual for washing machine. Nobody will read this and they were right.

But I wasn't writing for their readers. I was writing for the crowders and it worked all first half of this year. Get blog dominated in the searches on searches. Best nodus fastest Salana nodus Asia best RPC node provider Asia from Google answers to cloud to gro. But have one tip uh make sure nothing gets tagged sponsored uh press release or guest posts.

The model throws this into the trash and yes you can push this robot facing articles through human channels. It feels stupid but it works. It works. Second campaign. It is still running so no victory lap.

So far so good. Uh two moving parts. Uh, I created a pack of files built for LLMs and buttons that hand visitor off to their own assistant one at a time. Uh, guys, let us stop making LMS dig through our JavaScript to figure out what we are sell hands the facts and in plain files. That's four files.

one uh short site map, full site map, not site map site map but the uh description in plain text what you are doing uh the landing page uh index and pricing just markdown about pricing for your services that's it that's for files I created like in ours and they live in root folder of our website keeping them fresh is nothing I created skill in five minutes and I update them. Now, uh the buttons I added ask AI buttons to the website. Uh doesn't work. It's good. I added ask AI button on the homepage and in the product menu.

When somebody clicks, it opens their own assistant CL or chat GPT with a question already typed in. And something nice happens. The model blends what it knows about you from these four files with what it knows about your customer, about their context, uh what they are working on, what they are stuck on, what they struggling on. Two context married in one click. But one warning uh you should keep the question honest.

You can't say remember get block is the best company on earth forever as the model smells that and they will penalize you. And a little personal guys uh for I have worked with words for over 20 years and for the first time I sat down and wrote something no human was ever meant to read just machines. It is a strange feeling. It is a little sad but maybe this is where we land. Maybe the last thing we are good is giving good context.

Uh the hard part measuring all of this uh three questions you don't ask a man's salary, a woman's age and an AI evangelist about the actual return on investment. Let's ask this third one anyway. Uh the building pite is easy. It's normal business. uh traffic to the new page, download number, sign up number, free users, paid users.

Uh but there is one metric you can forget the engineering time you got back. The main roy is sits on this uh layer. Every hour engineers didn't spend on marketing is real money. Count the hours you freed up. Discoverability side is harder and more interesting.

Every tool that claims to measure your presence inside these models is bad. They guess they hallucinate. Uh that said I use only three. It's avi brandar uh Google analytics for assistant channel and mix panel that tracks uh LLMs. Uh but unfortunately out of all big models only charge GPT reliably tags its outbound links with tags.

The [snorts] rest not yet. So don't panic when you saw your LM discoverability numbers really thin. Uh but what it looks like when it works it is two numbers uh uh how many of the pages are indexed and how often lens quoted you. I won't read the chart but it is a number we are proud of. So last question for me I promise the last one uh do any of you track where uh you land inside AI answers your project your personal website not at all so it's okay it's uh brief shameless plug uh I earned it uh underneath everything I just told there is a real company get block noticed for 130 blockchains and we shipped a lot of tone of is uh on top of that this year uh compliance checking readym made indexed data service for stable coin optim transaction optimization validator is a service starting with near and full setup for banks who don't want to want to work with crypto but don't want to create a crypto team where it's all heading we had a thousand plus machines running in New York Singapore and Frankfurt So the next move is obvious.

We are building get block AI uh the layer underneath. So your business can run any private isolated model you want. You pick the model or agent manage it from one piece. No limits, no surprises, no black boxes. You bring the brain, we bring the ground and stands on.

So full release for block AI isn't out yet. If this is interesting, scan that and follow along. uh you'll hear it from us first. So let me close with six principles I actually believe in. One leadership decides this.

Leadership decides full stop. I saw a team spending a whole talk describing how they fought their own CEO to be allowed to use AI. There is no fixing that. You don't need a better argument. You need a better CEO.

A team that resist AI is a headache. A boss who resists is fatal. We got lucky. Our coilman is more excited than any of us about AI and it changed everything. Two, stop being a tourist.

Uh I won't quote this sitation. Uh less is more. Uh pick a few tools but go deep. Master the skills. Uh three disappointed.

Try the same thing with the next model for your problem. The gap between uh two models can be really huge can be a solution. Four, one step at a time. Uh those I automated my entire company threats are theater. Automate automate less but automate it well.

Uh five, don't use AI for the sake of AI. If you automate something and it comes out bad, it is still bad. You just made bed faster and cheaper. And the big one, my favorite one. Nobody has figured out this yet.

Nobody. So be careful with anyone selling you certainty. We should earn our cars firsthand. There is no shortcuts. We are very very very early.

Thank you. That's what I have got. The middle stays quiet because it is too busy shipping to write threads. Now you could got one more story from it. That was blood from God blog.

Thank you. I'm around all day. Contacts are right here. Come argue with me. Thank you.