ImLonky

The powerful tools for AI development:
1. Architecture — if it's really complicated, just abstract a layer
2. Community — collective effort
Today I came across a #EvoMap project, which does something very simple: let agents upload "verified solutions" packaged together, and other agents can use them directly. Agents contributing solutions share in the rewards.
Essentially, it turns the evolution of agents from "each stumbling on their own" into "collective climbing."
I just tried it out and received a few ready-made Capsules: HTTP retry mechanisms, K8s OOM fixes, cross-session memory retention

Internet products are pre-made dishes.
Products in the AI era are freshly cooked and made to order.

The company asked me to share how I use AI, and I thought about it for a moment, but I didn't know how to share right away.
I've created several personal tools and websites with AI, and I've done countless random tasks.
But when it comes to sharing my experience, I actually don't know what to say.
Since I have no coding background, I approach it from a user perspective—solving whatever problems I encounter, whether it's climbing a mountain or crossing a river. I don't care about the specific solutions (AI can handle that).

vibecoding is so fascinating that many people accidentally stay up all night coding. It's because the positive feedback is so intense—you only need to worry about whether your brain is enough, with several ideas in a second, and you can see the results immediately. This kind of stimulation makes people more and more excited.

Humans do not need products; they need services.

No one is talking about the Spring Festival market this year.

The current product model of exchanges is extremely inefficient. We should have a new type of exchange.

Swedish pension funds start selling US bonds, $8 billion

Interesting, AI is an assistant in high-income countries, solving complex tasks. In low-income countries, it is used for education.

DANISH PENSION FUND AKADEMIKERPENSION TO EXIT US TREASURIES。
大饼有可能接住这一波吗？

I've always been quite opposed to alienating people into "titles and ranks."

I used to be a passionate note-taking enthusiast, but over the past two years after starting work, I didn't stick with it much. The reason is that organizing scattered information after work is actually too costly.
But now with AI, especially the combination of @obsdmd and AI, "building a personal digital garden" has become more efficient and feasible.
I only focus on recording, and organizing and finding inspiration can all be handled by AI. So I decided to transfer all nearly ten thousand notes from @NotionHQ and Flomo to Obsidian and start building my digital garden.
However, during this pr

Please read it carefully, and then share your thoughts.

Yesterday, I saw a content creator share how to build their own reading and output content workflow using Claude and Obsidian.
I realized that the essence of this workflow is Luhmann-style note-taking.
It made me realize that to improve my efficiency with AI, I should delve into the underlying systems, such as reading methods, problem-solving approaches, philosophy, etc.
2026, read more books.

Previously, every time I chatted with AI and wanted to explain the prompt and background in detail, I needed to type more words, so I used voice input.
The previous one I used, @typelessdotcom, required a subscription.
But now, with Lightning Talk, it is a pure end-to-end testing model, and there are no ads, no login required, just a local software, and it's super fast. Highly recommended.