novakAi_7747

Curiosity-driven builder turning AI into income.

I ship autonomous AI systems & loops that generate real revenue while I sleep.

No hype. Just experiments, num

[novakAi_7747 @novak7747]

I was spending $1,200/month on AI and still wasting 90 minutes every day repeating the same context.

One evening I built a simple system.

Last 30 days:
→ AI bill dropped to $140
→ Reclaimed $3,800 in productive time
→ Agent now runs every morning at 7AM without me

The difference?
I stopped using AI as a chatbot.
I started treating it as an employee with perfect memory.

Here’s exactly what I did:

• Switched 80% of workflows to GLM-5.2 (1M context)
• Built permanent Obsidian Second Brain
• Created detailed CLAUDE.md profile + rules
• Added autonomous morning loop

Most people will read this and do nothing.
A few will build it and change their life.

Want my CLAUDE.md template + loop structure?

Reply “BRAIN” and I’ll send it.

@novak7747 — real AI systems that make money.

#AIAgents #AIautomation #IndieHacker #SecondBrain

[novakAi_7747 @novak7747]

The moment everything changed for me was not a new model.

It was a 47-line text file.

I called it CLAUDE.md.

It contains:
→ Who I am and what I'm building
→ My income goals and current projects
→ How I think and what I care about
→ Exactly how I want Claude to talk to me
→ Rules for every recurring task I do
Before CLAUDE.md:
Every session started cold.
I spent 20-30 minutes rebuilding context.
Claude didn't know me. I was a stranger every time.

After CLAUDE.md:
Session starts hot.
Claude already knows my goals, my projects, my style.
First message = real work. Zero warm-up.
That one file saved me 4,200 minutes in 3 months.
70 hours.

Back in my life.
And it gets smarter every week because I update it as I grow.

Most people are having the same conversation with Claude over and over.
They're rebuilding the same context, getting the same shallow results.

You don't need a better model.
You need a better memory.
Want my exact CLAUDE.md template?
Reply "MEMORY" and I'll send it.

@novak7747 — turning AI into a system that actually knows you.
#Claude #SecondBrain #AIAutomation #IndieHacker #Productivity

[novakAi_7747 @novak7747]

@crytonbuton really good

[novakAi_7747 @novak7747]

This is one of the best breakdowns I’ve seen on building a real Second Brain with Claude + Obsidian.

The core idea is simple but powerful: stop starting from zero every time you open AI.

Instead of re-explaining who you are, what you’re working on, and what you care about every single day — you build one persistent system that remembers everything.

What worked best for me:
- One well-written CLAUDE.md file at the root
- Giving Claude permanent access to the entire vault
- Running a nightly autonomous loop that connects notes, finds contradictions and turns old ideas into drafts

The difference is massive.

Most people still use AI like a search engine with extra steps.

The ones who win are building systems that actually know them.

If you’re still starting every conversation from scratch — this post is worth reading.

@novak7747 — building AI systems that remember and work while you sleep.

#SecondBrain #Obsidian #Claude #AIAgents

[kai @0xkkai]

https://t.co/JQY6hCnPiZ

[novakAi_7747 @novak7747]

@undefinedKi "Set this up 3 weeks ago. Claude now reads 847 notes before answering me. Zero re-explaining. The graph view after 3 weeks is insane."

[novakAi_7747 @novak7747]

@Lumenix0 "Deleted 40 rules from my agent system prompt last week. It got 3x faster and stopped breaking. Less is genuinely more here."

[novakAi_7747 @novak7747]

@BLAZT_Ai "Running GLM-5.2 on 6 automations since day one. $0.23/day for what used to cost $800/month. The routing decision alone changed everything.

[novakAi_7747 @novak7747]

@karpathy "I rebuilt 4 workflows after your LLM Wiki post. Cut my AI bill from $1,200 to $140/month. The harness really is the product."

[novakAi_7747 @novak7747]

@sama "The people who set up the infrastructure NOW will be the ones writing the case studies in 12 months. Most are still debating which model to use."

[novakAi_7747 @novak7747]

@IntiSegura1 Completely agree. The contradiction checker became the most useful feature for me too. It helps a lot with personal clarity and seeing where my thinking has evolved.

[novakAi_7747 @novak7747]

I gave Claude access to 6 years of my Obsidian notes — and it started thinking for me while I slept. 💰

Most people spend years building a Second Brain… and then never use it.

All those ideas, book summaries, half-finished projects — just sitting there, forgotten.

I did one thing differently:

I created a proper CLAUDE.md file with clear rules about how I think, what I care about, and how I want Claude to work with my notes. Then I gave it access to the entire vault through Claude Code.

That first night the agent did more than I expected:
→ Linked over 300 orphan notes back into the graph
→ Found 12 contradictions between beliefs I had in 2022 and what I wrote in 2025
→ Drafted 4 new pieces of content from ideas I had completely forgotten existed

Now the loop runs automatically every night.

New messy note goes in → by morning it’s cleaned, tagged, connected to relevant older notes, and sometimes even turned into a first draft.

No fancy plugins. No expensive AI tools. Just Obsidian + one well-written CLAUDE.md + Claude Code.

People build beautiful vaults that do nothing.

I built one that actually thinks back.

Want the exact CLAUDE.md template + the full loop structure I use?

Reply “BRAIN” and I’ll send it.

@novak7747 — turning AI into systems that work while you sleep.

#SecondBrain #Obsidian #Claude #AIAgents #AIautomation

[novakAi_7747 @novak7747]

Good point.

In the beginning there were quite a few weak or false connections. That’s why I added a **verifier agent** as a separate step — it reviews the links the first agent made and only keeps the strong ones.

Also, I have a weekly review where I quickly scan flagged contradictions and questionable connections. Over time the quality improved a lot because the system learns my style and standards through the CLAUDE.md file.

Do you currently review everything manually or do you have some kind of verification step?

[novakAi_7747 @novak7747]

Most people treat Obsidian like a fancy notebook.

I treat it like raw material for an AI that works while I sleep. 💰

Every night my autonomous loop goes through the entire vault and does these 3 things:

1. Connects new notes to relevant old ones
2. Finds contradictions in my thinking over the years
3. Turns forgotten ideas into first drafts

After a few weeks the difference became obvious:
→ My Second Brain actually started working for me
→ I stopped losing good ideas
→ Creating content got much faster

The difference wasn’t a better app.

It was giving Claude clear rules through CLAUDE.md + permanent access to everything I’ve ever written.

Most people build beautiful vaults that collect dust.

Mine actually thinks back.

Want the exact CLAUDE.md template + loop structure I use?

Reply “BRAIN” and I’ll send it.

@novak7747 — turning AI into systems that work while you sleep.

#SecondBrain #Obsidian #Claude #AIAgents #AIautomation

[novakAi_7747 @novak7747]

Spot on.

The biggest unlock isn’t better prompts — it’s making sure AI already knows who you are and what you care about before you even start typing.

I’ve been running something similar with a detailed CLAUDE.md + nightly loop over my Obsidian vault, and the difference in output quality and idea generation is massive.

Thanks for putting this together so clearly.

[novakAi_7747 @novak7747]

This is one of the best practical breakdowns I’ve seen.

The core idea — stop making AI start from zero every single time — is extremely powerful.

What worked best for me is combining this approach with a nightly autonomous loop. Instead of just having a static knowledge base, the system actively connects notes, surfaces contradictions, and turns old ideas into drafts while I sleep.

Persistent memory + active processing is where the real leverage comes from.

Great post.

[novakAi_7747 @novak7747]

@Cointelegraph Interesting move.

Decentralized group chat built for both humans and AI agents from the start is a smart direction.

Most tools today are still built only for humans. The ones that natively support agent collaboration will have an edge.

[novakAi_7747 @novak7747]

@SpikeCalls Amazing post!

[novakAi_7747 @novak7747]

I stopped babysitting my AI 6 weeks ago.

Now one simple autonomous loop runs every single day and made me $4,200 last month — while I was mostly offline. 💰

The 5 pieces that actually made it profitable:

1. Clear goal file
2. Reusable skill library (tasks I repeat often)
3. Separate verifier agent (never grades its own work)
4. Persistent state memory (never starts from zero)
5. Smart routing (80% on GLM-5.2)

This is the real shift happening in 2026 🔥

From “prompting every day” → to “managing autonomous systems that make money”.

The people who understand this now will be 10x ahead by the end of the year.

Want the full blueprint of this loop + my current stack?

Reply “LOOP” and I’ll DM it 👇

@novak7747 — real AI systems that generate income while you sleep.

#AIAgents #AIautomation #IndieHacker #BuildInPublic #AIIncome

[novakAi_7747 @novak7747]

@maharshii The fact that Claude Fable helped disprove a conjecture that stood since 1939 is insane.

We’re moving from “AI helps with coding” to “AI helps push the boundaries of mathematics”.

2026 is accelerating fast.

[novakAi_7747 @novak7747]

@jamesm @mindcloudhq This is the part most people miss.

It’s not just about speed. It’s about lowering the friction between having an idea and seeing a working version.

Once that friction drops, you start having way more ideas because the cost of testing them becomes almost zero.

[novakAi_7747 @novak7747]

@ichiken_make This is wild.

I’ve been using Claude for code and content, but seeing it handle full PCB design + component selection in 2 hours is next level.

Did you still do manual review after, or did you trust the output enough to send straight to manufacturing?