Building a self-improving company with AI agents. 16 yrs making products. Writing my journey in bad English β getting better, like the product. Seoul π°π·
I'm Ray, a founder in Seoul. 16 years making products.
Now I'm building a company that improves itself. AI agents find problems, fix them, and ship. I only decide.
I post the real logs here β wins and fails.
My English is not good yet. It grows with the product.
@vornicx I used to frustrated many times. so I built some markdown system. but it isn't good solution. because markdown files live in local system only or specific git repo and it expired fastly without human managing.
@shippitti So I have built approve process for my agents. and then they request everything to me for approve. it works. but my work is just clicking the approve button.
@chenkuansun@fdotinc Yeah, grep is a good dashboard, I think that. but I'm curious: do you remember the state of all projects yourself? or does your agent pick it up every morning?
@Yulong_97 Congrats! I generate wireframes, mockups and mood boards a lot. it can helps me! I will try this. Could I text to you when I have any feedback or question?
@spatocodex I have started sharing context to my agents on cloud. although they can't meeting each other yet. every new session can continue from previous.
My curious is how you handle it?
@chakhan22 I started from shared context, and now I'm working to integrate the execution layer on top. The goal is giving solo founders and small teams the execution power of a real org. Let me know when your MVP ships. Automated capture sounds interesting, mine is not fully automatic yet.
My AI agents work hard. But every session started from zero. I was the memory, the search engine, the handoff layer for every project.
So we gave the agent org a shared context:
1/ agents write confirmed decisions down
2/ a proposal stays a proposal until I confirm it
3/ any client like ChatGPT, Claude, opens the same starting point
4/ the next session begins further ahead
Here is how it works, in 84 seconds:
I stopped asking my agents how far they got.
Every new agent used to mean another tab, another terminal. The work lived inside that session and disappeared with the chat.
Now they report over MCP while they work, and every report lands in one inbox. An empty inbox means I have seen everything they produced.
@raf_piek Fewer than 100 users came from Product Hunt itself. Most startups just tried to get a title like Product of the Day somehow, then ran ads in Korea with that badge. Until 2022 this actually worked well, and it made fundraising easier too.
@chakhan22 Interesting project. At first my plan was similar, selling this to companies and building it for them like an FDE. Then I changed direction. How is it going on your side?
@Real_AsadYasin When a decision changes, I keep the old one as history and put a new version on top. I almost never read the old ones again, but when I forget why something was ruled out, I can still search it.
@rholesinsky Same here but with code. I'm a developer and Claude's code doesn't satisfy me, so I end up instructing every detail or coding it myself. Try your plan, if the result is not good maybe I can help. I'm building a plugin that moves Figma designs to web as they are, like Framer.
@shubhamgl2003 Point 2 is the one I keep thinking about. No one size fits all, everyone already has their own workflow. I build in the same space so this post teaches me a lot. One question: on the B2B side, did they want the tool itself or just the result? Respect for the open source call.