Twitxy Saved

Nikolai Pashuk

Verified account This account is verified. Learn more

@pashuk

Follow to learn how to do your marketing with AI.

Building one-person marketing departments.

🛠️ github.com/adaptico/adapt…
🏗️ adaptico.digital
adaptico.digital Joined April 2009
406 Following 1.4K Followers 410 Posts

Latest posts 5

Pinned post

Nikolai Pashuk Verified account @pashuk
I haven't posted for over a month. It was burnout, and I'll write about it tomorrow. But I'm back now.

I'll be here a lot more from now on, so here's who I am and what I'm building.

I'm a marketer and founder. I've spent a decade selling things on the internet:
> Launched simple web projects since school and marketed every one of them myself
> Found the niche that paid: CPA networks. Years of performance marketing on my own money
> Sold every offer you've seen online, from enlargement pills/creams/pumps for a certain organ (yes, that was me pushing them on you) to bank loans
> 2017: cofounded my first project with a team - 25 people. I ran marketing and some other things. The project failed: the market moved while we were building it. (The photo is from that year - the hackathon we won. Narrator: everyone is smiling here. They didn't know yet how it would look a year later.)
> Then a marketing agency for early-stage projects, and CMO or fractional CMO at several of them - building the whole marketing function from nothing
> 2025/2026: joined the founding teams of two projects (post-mortems will follow). One reached the top 10 on its chain in about half a year, then couldn't raise. The other closed early when the market changed

When those two shut down, I understood I'd been making the same mistake all that time.

I was always behind the scenes, so nothing carried over. No audience, nothing with my name on it.

Now I'm building Adaptico OS in public - an open-source go-to-market operating system, 28 commands so far, on GitHub. It's one of a few things I'm working on - the others I'll show when they're ready.

Here's what I'll be posting about:
> What changed in marketing since 2017 - and again since 2025 - and what stayed the same
> The set-and-forget AI marketing myths: what the agents really do, and where they still need you
> The exact tools I use for each marketing job, instead of one chatbot for everything
> The launch mistakes I've watched kill projects
> The real numbers of building all this in public - wins and flops

My goal for the rest of 2026 on X: show how one person can actually run a project's whole marketing with AI - no "my agents do everything" promises, just the things that work - on live projects.

If that's the kind of thing you want to see more of, follow me. I'll be posting all of it here.
Nikolai Pashuk Verified account @pashuk
Every feature you build and every direction you take comes from your research.

If the research is bad, you spend months building a SaaS nobody needs.

To make mine better, I run the same question through several AI tools, because each one reads different sources:

1. Write down the one question you need answered.
(for example: are there customers already paying to solve the problem you want to build and launch for, and how much?)

2. Run it through at least two of these:
> Gemini Deep Research
> ChatGPT Deep Research
> Claude Research
> Grok, which also reads what users post on X
(if you only pay for one, ask a friend who pays for another to run it for you)

3. Save each report in one research folder.
(if you use Claude Code or Codex, ask it to organise the reports for you)

4. Give that folder to a separate AI chat or agent and ask it to merge the reports:
"Mark what the reports agree on and where each claim came from. List where they disagree and keep both sides. Flag anything with no source."
(ideally, save this as a skill so the merge runs the same way every time, but it works without one too)

5. Open the sources behind every disagreement yourself.
(this is where you stay in the loop: agents gather and merge, you check)

6. Open the sources behind the agreements you're about to rely on.
(all of them can quote the same blog post, or a source you wouldn't trust)

7. Write down what you decided and why, in the same folder.
(your next round of research starts from there)

Use it before your next launch, or any other big decision about the product.
Nikolai Pashuk Verified account @pashuk
Vibe coding is the best thing that happened to solo founders and indie hackers: a feature for your SaaS takes an hour now.

Here's what the feature really costs you in addition to this hour:

The users you didn't talk to
> While you were building, you weren't working on your go-to-market: finding customers and talking to them. If you just launched, you could have been asking the first visitors why they didn't sign up.

A later launch
> If you haven't launched yet, every feature pushes the launch back, and with it the day you find out whether anyone wants the product.

The work around the code
> 37signals, the company behind Basecamp, listed 23 steps for adding one feature, and writing the code was step 6. A solo builder skips some of them, but you still have to test the feature, write the help page and update the landing page. And one feature leads to more: team accounts need invites, roles and billing.

The bugs
> CodeRabbit, a code review tool, checked 470 code changes in open-source projects. The ones co-written with AI had about 1.7 times more problems than the ones written without it. A vibe coder working alone has nobody to hand the bugs to, and each fix and review is another hour not spent on distribution, getting the product in front of customers.

A product that takes longer to understand
> A visitor gives your landing page a few seconds. Each feature you add to the page makes it a bit harder to see what the product is for in that time, and some visitors leave before they do.

A product that's harder to learn
> A study in the Journal of Marketing Research concluded that once a product has enough features, one more shouldn't be added even if it costs nothing to add. The authors called it feature fatigue: each feature is one more thing a new user has to learn.

And after all that, the feature may sit unused. Pendo, an analytics company, looked at 615 software products. 12% of the features got 80% of the daily use. 80% of the features were rarely or never used.

Why founders and builders keep building features instead of finding customers is in the article below, with a 7-day plan to stop.
Nikolai Pashuk Verified account @pashuk 3d
View quoted post on X
12 replies 0 reposts 8 likes 347 views View on X
Nikolai Pashuk Verified account @pashuk
If you're a solo founder running your SaaS marketing with AI, every answer is a guess until it knows your product.

Everyone running AI agents has a soul.md, a user.md and an agents.md by now. They tell the model who it is, who you are and how to work. They say nothing about your product, your users, your rivals or what you already tried.

That's the go-to-market part most founders skip. Here's what goes wrong, and what fixes it:

It gives advice that fits any product
-> product.md: who it's for, what they use today, why they'd switch

It tells you to try every distribution channel at once, with a team you don't have
-> stage.md: where the project is (still building, just launched, first users), the next step, your budget, your hours, anything else you can't do

It doesn't know how your users talk
-> audience.md: where they talk online, and what they say there, word for word

It has never looked at your competitors
-> competitors.md: their pages, prices and what their users complain about

It suggests what already failed
-> log.md: what you tried (a Product Hunt launch, cold email, building in public), what came of it, and what you ruled out

It has no idea what your numbers are
-> no file: connect your analytics and payments through MCP or an API, so it reads your signups and paying customers live

It agrees with every idea you have
-> one line in agents.md: "list what's wrong before what's right"

It repeats a mistake you corrected yesterday
-> rules.md: every correction, written as a rule it reads next time

It writes slop
-> examples/: if you publish anything, 10 things you wrote and 10 from others that made you stop and read

Each file only needs what the model can't guess about your product.

You can set up a first version in a weekend. After that you stop explaining your product in every new marketing chat, and you spend your time correcting the model when it's wrong.