AI & GEO Journal Feb 2, 2026 · 3 min read

I Gave an AI Agent My SSH Keys and Let It Code on a Live Site.

Three dev tasks worth a $3,000 agency quote, done by an open-source agent for $76 — at about 85% quality.

I Gave an AI Agent My SSH Keys and Let It Code on a Live Site.

The short version

OpenClaw, an open-source agent built by a solo developer in Austria, hit 100K GitHub stars in two months and handled three real development tasks on a live test site for $76. The output was roughly 85% right, and the real signal is that the pattern arrived from open source before any commerce vendor shipped it.

I didn’t expect the future of eCommerce operations to come from open-source.

But here we are.

Meet OpenClaw

OpenClaw is an open-source AI agent that runs 24/7 on your computer. You message it on WhatsApp, Slack, or Telegram like a coworker — and it actually does things. Browses websites, writes code, manages files, sends emails, runs scripts. All on its own.

Think of it as an employee who never sleeps, never takes breaks, and follows instructions literally. For better and worse.

It hit 100K GitHub stars in two months. CNBC, Wired, and Forbes covered it. And it was built by a solo developer in Austria.

Not Salesforce. Not Adobe. Not SAP.

That part I didn’t expect.

Testing Pinchy

I installed it a week ago. Named it Pinchy. (If you know, you know.)

And I had to test Pinchy on something real.

So I pointed it at a test website and gave it three tasks: add Google Places autocomplete to a form, build a coupon code feature, and fix a handful of bugs.

Pinchy worked through them autonomously. I messaged it on WhatsApp, it SSH’d into the server and wrote the code.

Is the output perfect? No. Maybe 85%. I still needed to review and clean things up.

But those three tasks would’ve been a $2,000-$3,000 agency quote. Minimum two-week turnaround. Back-and-forth on specs. A project manager who adds no value.

Pinchy did it in hours for $76.

Why eCommerce Leaders Should Pay Attention

Now scale that thinking to a distributor managing 15,000 SKUs.

Imagine pointing an agent at your product data and saying “find every inconsistency between Akeneo and Shopify Plus.” Or “monitor these five competitor sites and flag price changes on key SKUs daily.”

That’s where this is going.

The newsletter · Twice a month

Get the next one before LinkedIn does.

Real B2B commerce implementations. PIM, ERP and eCommerce integrations that work. What to do Monday, in your inbox.

2,050+ operators. Unsubscribe in one click.

The Honest Limitations

We’re not there yet.

No native eCommerce integrations. No Shopify connector, no PIM hooks. Setup requires real technical skills — terminal commands, configuration files, API keys. Most eCommerce teams I’ve worked with at distributors couldn’t install this without developer support.

Security needs serious attention. OpenClaw gets full machine access. Hundreds of exposed instances were already found on Shodan within weeks of launch. If you handle customer data or payment info — and every distributor does — you need a sandboxed environment. Not your daily work laptop.

And API costs at scale are a conversation nobody’s having. Running Claude Opus 24/7 on complex tasks burns through tokens. That monthly bill adds up fast.

Why I’m Excited Anyway

The pattern is unmistakable.

Two years ago I was duct-taping AI content enrichment tools together at a $30M distributor. It felt experimental. We had to build everything from scratch.

Pinchy is what happens when that duct tape starts becoming infrastructure. Autonomous agents that monitor competitors, clean product data, and execute operational tasks without constant hand-holding.

And the fact that it came from open-source before any enterprise vendor got there? That should worry every VP of Digital who’s waiting for their platform provider to “add AI.”

They’ll be late. Again.

What This Means For You

You don’t need to install OpenClaw tomorrow. But you need to understand what it represents.

In 18 months, your competitors will have agents like this embedded in their operations. The distributors that start experimenting now — even imperfectly, even at 85% accuracy — will have a massive head start over those waiting for a polished enterprise solution that may never come.

The future of B2B eCommerce operations just got a preview.

Start paying attention.

Questions I get on this

FAQ

What did the agent actually do?

Added Google Places autocomplete to a form, built a coupon code feature and fixed a handful of bugs — instructed over WhatsApp, executed over SSH on the server.

Can a distribution team use this today?

Not directly. There are no native eCommerce integrations, no Shopify connector and no PIM hooks, and setup needs real technical skills.

What is the security risk?

The agent gets full machine access, and hundreds of exposed instances turned up on Shodan within weeks of launch. Anything touching customer or payment data needs a sandboxed environment, not a daily work laptop.

Rudy Abitbol

Moved from L'Oréal and Club Med campaigns to running B2B eCommerce at Sonepar and a $30M medical distributor. Now advises $5M–$1B distributors and manufacturers on product data, Shopify B2B and AI search. Montréal.

Also published on the newsletter

AI & GEO

Want this built with your team?

What AI engines say about you today, and what makes them say something better.

See the service →

Scope, timeline and what your team owns at the end.