NewsletterFebruary 18, 20265 min read

I have been living in the future

Two weeks with OpenClaw and Claude Opus 4.5 changed how I work. From vibe coding to managing a team.


Table of Contents

Two weeks ago, something exploded on Twitter! An open-source project called Clawdbot — an AI agent that runs on your machine. Not a chatbot. An agent that actually does things.

I was already deep in the AI building world. I'd been vibe coding for months. I'd even built my own AI agents inside Temaki, the workspace platform I'm developing. But Clawdbot was different. It wasn't just a tool — it had personality. And it did everything I'd been trying to make my own agents do.

So I set it up. And within days, something shifted. I wasn't vibe coding anymore. I was finally managing a team.

From vibe coding to managing a team

There's a difference between asking AI to write code and running a team of AI agents that plan, execute, and coordinate with each other.

I built mine fast. Each agent has a name, a role, a personality, and a domain. One handles strategy. One handles content. One handles security. They talk to me through Telegram. They talk to each other. They remember things between sessions.

This isn't vibe coding. This is something new. I'm not writing prompts — I'm setting a vision and building a team to deliver it. I'm reviewing work. I'm making decisions while my team executes. The mental model shifted from "me + AI tool" to "me + team."

Coincidentally, that's exactly what I do at my real job.

The model that changed everything

But Clawdbot was only half the equation. The other half was what was running behind it: Claude Opus 4.5.

My agents started writing copy that didn't sound like AI. They made good design decisions. They pushed back on bad ideas. They had opinions.

It wasn't a speed improvement or a bigger context window. It was personality. Taste. Judgment.

Opus 4.5 didn't just code better — it thought better. It understood what I was trying to achieve, not just what I asked for.

That's why my personal website finally worked after months of failed attempts. It wasn't just Claude following instructions. It was Claude making editorial choices — layout, typography, tone — that felt authored. Not generated. Authored.

The combination of Clawdbot (the framework for managing agents) and Opus 4.5 (the intelligence behind them) is what turned everything upside down. One gave me the team. The other gave the team a brain worth managing.

And the output changed dramatically.

What I built in two weeks

I started more new things in two weeks than what I am ready to admit. I'll share what I published.

A personal website I'd been failing to build for months.

Every few weeks, I'd try again. Cursor, Lovable, GPT, Gemini — I tried them all. Every time I got the same thing: clean, functional, completely soulless. The "AI look." I'd get frustrated, close the laptop, and give up. Not because the tech couldn't do it, but because it couldn't do what I envisioned. It had no taste.

This time, using Claude Opus 4.6 in Claude Code, I finally got it. Something opinionated. Editorial. Bold. Not a template with my name on it — a site that feels like me.

I didn't touch a single line of code. I briefed, gave feedback a few times, and then I hit publish.

fernandoamaral.org

A browser-based flight simulator. In a week.

Photorealistic cities. No downloads. Just open a URL and fly over any city in the world. Also built without writing code.

worldflightsim.com

A year ago, I spent weeks getting a content writing assistant to work properly. Now I'm shipping flight simulators before dinner.

The Clawdbot story

This tool is central to everything I am doing and you should know its story — because it's one of the wildest rides in tech.

November 2025: Peter Steinberger, an Austrian developer known for building PSPDFKit, publishes a side project called Clawdbot. An AI agent that runs locally, connects to Claude or GPT, and does real things through your messaging apps. Not a chatbot. An agent.

January 2026: It goes viral. Anthropic sends a trademark complaint (too close to "Claude"), so he renames it to Moltbot. Three days later, he changes it again to OpenClaw because "Moltbot never quite rolled off the tongue." Meanwhile, someone launches Moltbook — a social network exclusively for AI agents. The internet loses its mind.

The numbers: 145,000+ GitHub stars. 20,000+ forks. Fastest-growing project in GitHub history. TechCrunch, CNBC, Wired, PCMag all cover it. Silicon Valley adopts it. Chinese developers fork it for DeepSeek. Peter does Lex Fridman's podcast (#491 — 3+ hours). He reveals both OpenAI and Meta made acquisition offers.

February 2026: ClawCon happens in San Francisco. The community goes wild.

And then this happened

Sam Altman just announced that Peter Steinberger is joining OpenAI to drive the next generation of personal agents. OpenClaw will live on as a foundation-backed open-source project that OpenAI will continue to support.

Altman's exact words: "The future is going to be extremely multi-agent and it's important to us to support open source as part of that."

I've been living in that multi-agent future for two weeks. It's real. And it's about to get a lot more resources behind it.

It's also happening in Lisbon. There's an OpenClaw meetup coming up — hands-on walkthrough of the architecture, multi-agent setups, and real-world use cases. 150 people registered, already sold out, but you can still join the waiting list. I'll be there.

Waitlist: luma.com/76kcxp2a

The world has changed

The world has changed, and it will continue to change.

The gap between having an idea and shipping it is nearly gone. The gap between people who build with AI and people who don't is becoming a canyon. Small teams, or even one person with the right agents, can now do what used to require departments.

AI can write code. We knew that. But now it has taste. It has judgment. It can make design choices that delight you. And when you stop using it as a tool and start managing it as a team, the ceiling goes through the roof.

This is week three for me. I can't imagine where this will be in six months.

That's it for this first issue. I'll be sharing what I'm building, thinking, and learning every week.

If this resonated, subscribe at fernandoamaral.org and share it with someone who should be paying attention.

— Fernando