There are so many stories saying, "It can be automated with just one click."
But when you actually try to apply it to your own work, it's easy to give up.
We've captured the raw records of a team that spent 6 weeks narrowing the gap between 'it works' and 'I'm actually using this for my work.'
The latest version of the experiments Selfish Club has been conducting internally for the past two years, the 8 members of Team AAA (AI Agent AZA).
Gathering every Sunday, we used Claude Code to directly integrate AI agents into our respective tasks. Some things worked, some didn't, and there were even days when we overturned decisions deemed impossible in meetings after discussing them with Claude.
What got stuck, what went wrong, and what we eventually broke through — we’ve captured it all exactly as it happened.
Why this team was created and how they worked together — before hearing the stories of the 8 members,
this is the intro part that covers the overall framework of Team AAA.
With the emergence of Claude Code, both non-developers and developers can now truly cross the boundaries of their roles.
However, there is a world of difference between "it only takes a click" and "applying it to my life and work."
Learning, applying, and truly making the flood of information your own
is too much and moving too fast to do alone.
So, instead of going solo, we decided to tackle it as a team.
That's how the team AAA (AI Agent AZA) was formed.
Eight people gathered every Sunday for six weeks to share and exchange what they had created.
It is not a grand product.
The story of two people that started from the work I do every day,
and the points where I always got stuck.
The goal of this part is to make you think, "I could try doing it this way too."
At first, I created a system where entering a single URL would simultaneously generate posts for a blog, Instagram, and LinkedIn.
The result was a failure. The tone didn't match, so I ended up working in Canva again, and it actually took more time.
So I changed direction.
The process of directly building a carousel editor, a simultaneous Threads/LinkedIn generator, and a system that even creates videos from the terminal.
It covers why automation failed, why I came back, and the design rules I structured along the way.
In an era where AI provides all the answers, I felt it was a shame that what "my own thoughts" actually are was becoming blurred.
If you drop a single line into Telegram, the AI asks you questions,
the conversation is automatically archived in a wiki, and
it becomes a content draft three days later.
I will cover the actual process I’ve been running on my own, however loosely, exactly as it is.
If I built an OS for myself, the next step is the team.
This is the story of turning scattered Notion pages into "living assets" that AI can read.
It was a waste to see the great records created weekly by Team AAA just lying dormant in their individual Notion pages.
As assignments and meeting records submitted weekly via Obsidian accumulated as data,
AI became able to extract and show what happened within the team and even the growth trajectory of each individual.
Since it is accumulated in the format that AI reads best (Markdown),
the archive is no longer just for "storage" but has become a "living asset" that is constantly retrieved and used,
and we cover everything up to an archiving system that can be shared externally.
"Can you really build something without knowing how to code?" This part is the proof.
Four real-world products started to solve different problems!
It covers the stories of sharing insights and feedback while exchanging processes and results through weekly missions.
I struggled with AI for three weeks and wandered aimlessly. I tried to follow "what others were doing" and failed at everything.
The moment it became clear "whose problem I was going to solve and what that problem was,"
I launched a real service called Jjim Market in just 4 hours by running 12 agents.
A record of how customer inquiries started coming in via KakaoTalk in just one day,
how to use Claude Code as 'your own co-founder,'
and a generous sharing of real-life experiences on how AI can go astray when the 'why' is missing.
No more spending hours every day jumping between multiple SaaS tools to collect data and organize performance results.
This covers how to check payments, reservations, CRM, and advertising performance all at once with a single custom dashboard built by integrating APIs,
and the reason why a marketer was able to build it themselves.
How did a marketer, who had been working as a PM without any development skills, come to lead an AX project?
She personally built an AX Dashboard to manage the entire project at a glance using Claude code, and
established an environment for visualizing information and
making immediate decisions by utilizing a Brainstorming Companion that keeps the project on track.
DB collection, application/payment pages, and even a member system!
From products created alone without a developer to actual payments! We cover the entire process.
9 types of notification talks + N channels sent every time a sharing session is held.
It takes 3 hours every time just to write copy, extract the database, and attach UTMs.
At first, I tried building it with Claude's code, but there was a problem.
If my MacBook turned off, the messages wouldn't send. That's where the story of moving to a server began.
I'll cover everything from how I created an automation where sending is completed with a single approval, to the actual stories of my failures.
Eight people each made their own, but in reality, not a single thing was done alone.
This is the answer that an overwhelming majority of the 170 pre-survey respondents cited as the biggest hurdle.
It also appeared repeatedly in the indirect expression, "There is no one to ask when I get stuck."
Every Sunday, there was an environment where eight people gathered to share their progress and provide feedback. At moments when I would have stopped and wondered, "Is this right?" if I were alone, a single comment from someone next to me saying, "This is how I did it," kept me moving forward again.
Gemma threw out a single suggestion: "What if we try using Obsidian instead of Notion?" With that one comment, Dada completely overhauled the team's recording method, converted it into data based on Obsidian + GitHub, and even went as far as creating a team archiving website.
There was a day when Owen saw Heumin's recording system and said, "I need to structure my thoughts too." Since that day, Owen started creating his own recording structure, which eventually even influenced the customer service flow of Jjim Market.
If I had done it alone, I would have quit halfway through. I also would have thought that my way was the only way to utilize AI. Although we were solving different problems, sharing our progress every week created unexpected connections, and that pushed each of our results to a completely different level.
People who have tried various things with ChatGPT but still don't have the sense that they have actually "built something."
The pressure of being a marketer who feels the need to cross over into development, or a designer in a situation where they must also consider product management.
People who have ideas piled up but are stuck at "how do I start."
A practitioner who can deliberate on AI and business together.
It is a community that expands your potential through AI. Marketers, designers, and entrepreneurs experiment with and learn AI together, practicing a way of growing by sharing knowledge instead of just accumulating it.
This is Selfish Club's flagship program, where AI practitioners share their firsthand trial-and-error experiences and achievements. It is not just sharing for the sake of learning, but a form of sharing where the sharer grows the most, allowing participants to gain insights that can be immediately applied to practical work.
Please add the Kakao Channel [Selfish Club] to inquire immediately. You can also contact us via email (public.selfishclub@gmail.com).
Selfish Club AX PROJECT — The raw records of 8 members of Team AAA
AI Agent AZA · A course containing the full 6-week experiment.
Thank you.
![[Free] Notion MCP: From Beginner to ApplicationCourse Thumbnail](https://cdn.inflearn.com/public/files/courses/338906/cover/01k4nkjw13vgrmrs0esxnc60w9?w=420)
