The Tech Nation — Newsletter #3

People. Tech. Impact.

Every 2 weeks. No fluff.

Article 01

Harness Engineering: The AI Skill Nobody Is Talking About

Nobody talks about it yet. But it's already separating winners from losers.

It's called harness engineering.

 
We spent two years obsessing over models. GPT. Claude. Gemini. Which one is smarter. Which one writes better.
That was the wrong race.
In February 2026, OpenAI engineers built a production app. Over a million lines of code. Zero written manually.
The breakthrough wasn't the AI model they used. It was the system they built around it.
They called it a harness.

 
Here's the simplest way to think about it.

Simple framing
The model is the engine.
The harness is everything else. The guardrails. The memory. The connections to your tools. The rules for when to ask a human before doing something stupid.
Without a harness, an AI agent is like a very smart intern with no manager. Capable. Unpredictable. Occasionally catastrophic.

 
Here's the stat that made engineers stop scrolling.

Benchmark shift
LangChain tested the exact same AI model twice. Same model. Different harness.
Performance jumped from 52.8% to 66.5% on a standard benchmark. No new AI. No new data. Just a better harness.
The model is becoming a commodity. The harness is where the real advantage lives.

 
So what does this mean if you're running a business?

Stop asking "which AI should we use?"
Start asking "how well have we built the system around it?"
Because right now, almost nobody in the boardroom is asking that second question. The ones who are, are quietly pulling ahead.

Article 02

Someone Gave an AI Its Own Computer. Things Got Interesting Fast.

In January 2026, a developer had a simple idea.
What if instead of asking an AI for answers, you gave it a computer and said: handle it.
Browse the web. Send emails. Book meetings. Execute tasks. No prompting. No babysitting.
He called it OpenClaw.

What happened next
Two weeks later: 200,000 GitHub stars. One of the fastest growing open source projects ever.
NVIDIA built on top of it. Tencent integrated it into WeChat. Then OpenAI hired its creator.

 
The core idea is almost stupidly simple.
A model on its own can only suggest things. Give it a machine to operate, and it can actually do things.
That combo, a capable AI model plus a hosted machine it controls, that's the new building block of automation. Not a chatbot. An operator.

 
And then the internet did what the internet does.
Within weeks, dozens of alternatives appeared. Built by random developers over weekends.

- NanoClaw rebuilt the whole thing in 500 lines of code. Cleaner, sandboxed, way harder to break.
- IronClaw rewrote it from scratch. Zero permissions by default. The agent can't touch anything without explicit sign-off.
- PicoClaw runs on a 10 dollar computer. Launched February 9th. Hit 12,000 stars in a week.
- Moltworker, built by Cloudflare, moved the whole thing to the cloud so the agent runs in an isolated box and can't go rogue on your actual machine. (Yes, that happened to real users with OpenClaw.)
Weeks. Not years.

 
Here's why this matters if you run a business.
Tools that used to take a team months to build are now appearing over a weekend, shipped by one developer with a GitHub account.
And the companies building the infrastructure underneath all of this, E2B, Browserbase, Daytona, they raised hundreds of millions of dollars specifically because AI agents need somewhere to live and operate.
This is a new computing category forming right in front of us.
Not AI that helps you think faster. AI that works while you sleep.
The question is not whether this goes mainstream. It already has.

Article 03

Crewly. We built the team we wish existed.

AI made building easier.
But it didn't make building properly easier.
Anyone can prototype today. Very few can ship something that actually works, scales and delivers value.
That gap is where most projects stall.
So people try to hire. And hit the same wall every time. Slow. Expensive. Uncertain.
We've seen it again and again. So instead of fighting the system, we changed the model.
We built Crewly.
A productized tech team. Not a freelancer. Not an agency. A real crew, ready to design, build and ship.
Product, tech, AI, delivery. Aligned from day one.
You don't assemble the team. You plug into it.

Discover Crewly →

Article 04

The Extras.

  Vibe Coding Masterclass
  Few spots left for this week. Very practical, no theory, real use cases.

  Register on Luma →

  AI Summit Barcelona
  Tickets are live. Get an extra 20% discount with the code :

  JEMW20

Article 05

Claude in Barcelona. Official. And growing.

We're doubling down.
After a first packed event, we are now officially hosting recurring Claude Code events in Barcelona, in collaboration with Anthropic and Happy Operators.
This is not just another meetup.
It's currently the only official Anthropic format in Barcelona. And one of the few in Europe with this level of direct connection to the ecosystem.
The ambition is simple.
Make Barcelona a serious node in the global AI conversation. Not just attending. Contributing.

Next edition

April 9 — World Trade Center Barcelona

Early Access →

Spots will go fast. They already did last time.

If this newsletter is useful, share it. If you're building something, let's talk.

You're building something. You want to move faster. Let's talk about how we can help.

LET'S TALK

People. Tech. Impact.