Meet Ed: Our Newest AI Teammate

Every week, a new AI tool promises to automate your business, replace your team, or at minimum, save you four hours a day. 

We've been building and testing agents for a while now — for clients, for ourselves, for products we're launching. Last month we wrote a newsletter about OpenClaw being a massive unlock for agents and predicted this would be the big story for 2026.

So this week, we thought we’d give an update on how that’s shaping up and share what (or should we say who) is actually working for us. 

His name is Ed.

Named for Edsger W. Dijkstra, the pioneering computer scientist best known for foundational algorithms and for shaping how modern programmers think about writing reliable code, Ed runs on OpenClaw – an open-source personal agent framework that has absolutely taken off in the developer world. You may have seen it making rounds online, because people are obsessed. There are even meetups for the OpenClaw community around the world.

In short, OpenClaw is a persistent AI that lives on a machine, connects to your tools and messaging apps, and takes actions autonomously.

When we started playing with it, we knew that if we set Ed up with the right tools (and the right guardrails), he would quickly become the best, most efficient member of our team. 

So we bought Ed a gift: a brand new Mac Mini. Then, we set him up with his own GitHub repo, Claude Code, and literally every tool the two of us use in our tech stack.

And once we did that, something clicked.

The most common way people think about AI agents is as a fancier version of a virtual assistant. Someone to handle scheduling, emails, maybe some research. That framing makes sense, until you see what these things can actually do.

We connected Ed to Claude Code, Anthropic's agentic coding tool, which runs directly on the machine. Now when we have a feature to build, a bug to track down, or a task that needs execution, Ed can pick it up, orchestrate Claude Code to handle the actual writing and testing, and come back when it's done or when it hits something it needs a human for. 

This is a fundamentally different kind of tool than a virtual assistant.

A good human virtual assistant runs somewhere around $2,000 a month, and that's for someone who works business hours, needs clear instructions, and can't spin up a code environment and push a commit.

We're not saying agents replace humans. They don't, not for most things that actually matter. But what they do offer is something no human assistant can: they're available constantly, they don't lose context between tasks, and once they're set up correctly, they can execute on their own.

The difference isn't "AI vs. human." It's "asynchronous, always-on execution" vs. "synchronous, availability-dependent help."

For certain categories of work, especially development, research, reporting, and content, the calculus is genuinely different now. Ed does work that would have taken us time we didn't have, on a timeline we didn't have to manage.

When we talk to clients about agents, we try to reframe the question. Most people ask: what tasks can I hand off? We think the better question is: what work is already repeatable enough that a system could own it?

That's the distinction between using an agent as a tool and deploying one as infrastructure.

Because the truth is, a great agent with a bad setup is just a fancy chatbot. The model is the easy part. What you do with the setup and the environment you give it to work in is everything. 

Ed is a member of the team, with a dedicated machine, a defined workspace, and clear ownership over specific parts of our workflow. When he needs us, he reaches out. When he doesn't, he keeps moving.

And by the way, if you’re interested in trying out your own setup with a Mac Mini, go get one today –  they're about to be in high demand. And no, we are unfortunately not getting kickbacks from Apple for this promotion.

Stay curious,

Julia & Russell