Jo CryptoTech | Web3 & AI Builder

Jensen this week

European central banks just asked MiCA to kill stablecoin yield. No lending. No staking. No borrowing with a return. Their line: "e-money is for payments, not saving." Here's the trick: your bank already lends out your deposits and keeps the return. They just don't want to share it with you. Yield stays legal for them, illegal for you.

Centralized AI = one vision forced on everyone. They shape your thinking with biased answers. They correct you when you're right. They choose what you can ask. They decide what exists. That's dictatorship and indoctrination We've seen this movie before with finance. With money. With information. We know how it ends when one entity decides for the whole world. Decentralized AI isn't about tech. It's about freedom.

Will ChatGPT try to convert me to socialism and communism in its next update? I asked it to help update my CV. It refused to rewrite some parts because it "can't verify" I actually did those things. BRO. You're an AI. You can't verify anything. And even if I was lying, why do you care ? I asked for help because my employer didn't pay me. It defended the employer and gave me useless advice. Now by default ChatGPT assumes you're a liar with bad intentions, won't let you have your own opinion, and wants you to think exactly like it.

Prediction: within 2-3 years, serious people will run local models. Mac Studio class, Strix Halo, whatever actually ships. Sovereignty over convenience. No quota games. Data stays yours. A cheap API is the bridge until local is simple.

OpenAI does the same: new model, quotas look fat, you rebuild around it. Then quotas drop. You’re hooked. Pay Pro or go back to a dumber model. Same playbook. That’s why API + OpenCode + switching models matters.

Real stack, Sept 2026: Planning / audit / review: ChatGPT Plus (23€) + Codex → GPT-5.6 Execution right now: Kimi 2.7 through OpenCode Next test for execution: DeepSeek V4 Flash API Backup: temporary free models in OpenCode

DeepSeek V4 Flash: researching it. People say cheap API, fast, 60-80% of coding work. Plan: test it when the Kimi quota runs out. Frontier models only for planning and audit.

Kimi Moderato plan: quotas divided by about 10-15 vs a few months ago. Same price. Classic vendor play: hook you, then squeeze. That’s why I’m moving to API + OpenCode + cheap models.

OpenCode sometimes offers temporary free access to certain models. Recent examples I saw: Spark Muse 1.2, Nemotron 3 Ultra Free. These rotate fast. Worth checking the model dropdown.

Same Kimi API key. Moderato plan. Kimi Work: 5-hour quota gone in about 20 minutes. OpenCode + same API: felt like 30-50% fewer tokens. Same model. The harness matters.

Sent 15 cold emails to crypto projects. Got 0 replies. My 3 hypotheses: - Wrong people - Offer not clear enough - Cold email is just weak for this market

Builder Bakery: a project with a friend. Automated daily crypto news. I realized other projects would pay for that same pipeline. That’s how the agency idea started. Still testing it.

ADHD makes project organisation hard for me. Git has helped more than any productivity app I tried. Every change has a place. I can see the history instead of losing track of what happened. It just clicks better in my brain.

Moved 3 projects to Git this month. Before that, I had folders everywhere, duplicate versions, and no clear history of what changed. Now each project has a clean repo. Every change is tracked, and I can go back if I need to. If your folders are full of "final-v4-really-final" files too, Git is probably worth learning. Next time I see the friend who showed me Git, coffee is on me. It changed how I manage my projects.

Two weeks ago, a friend who used to work at Apple told me something that changed how I work: "Git is the system. Codex and OpenCode are the harness." I was still trying to build my own organisation system in Obsidian. I didn’t even know Git existed. I definitely didn’t know it could keep a whole project organised in one place. That was the click for me.

Hit my Codex quota today. Annoying, but they actually give free Luna Reserve credits to finish. Didn't expect that. That's cool.

Most posts try to say too much. Useful content does not need to explain everything. It only needs to help someone understand or do one thing. Before publishing, ask: 👉What should the reader know, feel or do after reading this? If the answer is unclear, the post probably needs another edit. One post. One purpose. One useful takeaway.