Claude code, Kinen and AI generally

Waw it was such a crazy week, mainly driven by Claude code to be honest. I tried a bunch of new stuff and see if it can be useful in my day-to-day life, and it is.

I also started working on the new project that I will talk about more later.

Claude Code

After trying it a bit on some projects, I decided to use it more and pay for the Max plan to see all the possibilities.

As I wanted to build a brand new app, based on something I know with a lot of details, I used this framework that was very useful to me: https://github.com/bmad-code-org/BMAD-METHOD

It changed myself to a full manager, managing a bunch of agents doing the work for me. It's crazy as they are doing a way better job than before. When I was developing my own apps, I didn't write any test of stuff like that. I was mainly on the branch main, and that's all.

Now, they are writing a lot of tests, doing all the functionalities, helping me with brainstorming, and so on. That's crazy.

So I used it for this new project, and also now on a full recon dashboard that will help me in bug hunting. And I want to use it simply by using an LLM to scan what I want, putting a chrome headless to make test directly on the device and getting a get-around of an app before putting my hands into it. That will be great, and I will tell you more later if I got something nice.

Doing this app could have take me around a month before, debugging a lot of stuff, and now I can just chill on my couch and say "hm I don't like that, please change".

Kinen

On the project side, I’ve been thinking a lot about Kinen.

Kinen is the logical next step after Hackyx, but not in a “bigger Hackyx” way. More like: Hackyx was the first form of the idea, and Kinen is the version that’s honest about what I actually want.

With Hackyx, I felt stuck in a security-shaped box. The cyber angle is fun, and it’s my world, but it also narrows the product in ways I didn’t fully admit at the start. Especially the temptation of fully automated article ingestion: you can build pipelines that ingest and summarize everything… and end up with a knowledge base full of garbage.

The internet produces an infinite amount of content. If you automate ingestion without a strong filter, you don’t build a second brain; you build a landfill with a search bar.

So I reframed it.

Instead of “security news ingestion and automation,” Kinen becomes: an app to learn and manage a knowledge base. General-purpose, not cyber-focused. Something that helps me (and maybe others) deliberately collect, connect, review, and actually use information.

But it still starts from the core premise of Hackyx: structured knowledge, good retrieval, and workflows that support learning instead of just collecting.

I want Kinen to feel like:

• capturing things is easy
• organizing doesn’t become a second job
• context stays attached
• review is built-in
• the system pushes you toward understanding, not hoarding.

And honestly: this is where AI helps, but only if it’s shaped correctly. Summarization isn’t the goal. Insight and recall are the goal. I don’t want a machine that produces more text. I want a machine that reduces cognitive friction. I'm still working on it, but it will be up very soon, and I hope you will like it.

AI in general

It’s crazy the world we live in.

You can set up agents to work for you at night, doing the boring parts. The boring parts used to be most of the work. Now they’re optional, and you can forget about them.

The uncomfortable truth is that junior developers are being replaced. Not entirely. Not everywhere. But the “entry-level work” that used to be important, like CRUD, tests, refactors, boilerplate integrations, can be done within minutes with an LLM.

And it’s not just code output. If you give it a strong harness, good prompts, clear constraints, and a test suite, it can even find bugs. It can reason about edge cases.

But this only works if we learn to use it intelligently and safely.

Running agents in the cloud is powerful, but it’s also a trust and permissions problem.

If you give an agent access to your repos and your secrets, you are delegating authority.

So I’m trying to reduce the friction in my work but doing it securely. My agents don't have access to my emails, my documents, or stuff. I only give them what they need and are isolated.

But it's still crazy what you can do now. It's not perfect, but it's changing the world and the industry. What's next ?