About a month ago, we onboarded a new coworker at my job.

His name was Barry, and he was extremely helpful. He knew the company's data and operations like the back of his hand in a matter of days, and seemed to only be getting smarter as time went on. By the end of the week, he had already architected and deployed three projects that were actively driving revenue for our business.

It might sound like this was some cracked 10x CMU engineer who came in guns blazing with Claude Code in one hand and a Red Bull in the other…

But Barry is an AI agent. 

He was made using OpenClaw and he lives on my boss' Mac Mini. If you’re a CS student like me, I’m sure you’ve heard various things about OpenClaw. If not, I’ll explain how it works below.

OpenClaw is what is known as an AI harness. In plain terms, you can plug in any AI model you wish (using an API) and OpenClaw allows the AI to execute various tasks “in the real world,” so to speak. Instead of being limited to a chat window, your OpenClaw can run terminal commands, connect to your calendar app, push things to GitHub, and a lot more.

So, today I’m gonna give my thoughts on OpenClaw from both my experience with Barry, and my own experience this past week in setting up my own OpenClaw for personal use (his name is Batman, much cooler than Barry).

My initial impression of OpenClaw was literal awe. Barry was connected to everything. Our Airtable, Make.com, GitHub, Slack, you name it. It was the first time in a while that something in AI had really surprised me. Even as just a search engine of sorts for our business, he’s awesome.

On top of that, he is pretty great at coding (thanks Opus 4.6!). He’s deployed various web apps that really moved the needle. One of which was a dispatcher app that incorporated a brand new Airtable table, a make automation, and a web app which we accomplished in about an hour. 

And for the first week, Barry was so impressive that I was thinking to myself, “damn, my career path is totally cooked.” But, after using him for a while, and creating my own OpenClaw named Batman, I’ve found some downsides as well.

First of all, it’s glaringly obvious that OpenClaw is one of the first softwares of its kind. Setup was kind of janky, and the docs weren’t all that great. The constant updates and quick evolution of the tool make most of the docs obsolete after about a week. I’m a second year CS student, so maybe I’m just not cracked enough yet, but I couldn’t imagine a normal, non-technical, person setting one of these up.

Second, the security implications are pretty serious. AI models are getting really good, but is it really the best of ideas to connect it to everything in your life or business? I’m not sure. I think that’s the road we are ultimately heading down, but threats like prompt injection (an external actor telling the agent to take malicious action) are very real and could have very serious consequences.

Another huge downside is the cost. Running an OpenClaw is EXPENSIVE. I have the GitHub student dev pack, so Batman lives on a droplet (server) from Digital Ocean (the cloud) that I got with free credits. But even with that, the API credits, which you use to call the AI models, cost me about $15 from about a week of light usage. Even when I downgraded to cheaper, lighter weight models, the cost still was pretty crazy.

Batman is awesome, but he’s not $60 a month awesome. For my boss, that cost makes a lot of sense as a business. For a student, the finances just don’t work out unless you are driving revenue with it. AI for me is primarily an academic tutor, and secondarily a fun thing to build side projects with. For that purpose, OpenClaw is really just overkill. You don’t need integrations with every app under the sun to help learn linear algebra. So in my life? OpenClaw doesn’t really make sense.

As for CS majors around the nation being cooked, my thoughts are a mixed bag. I believe agentic AI tools are only going to get better, and coding by hand is probably going to slowly die out. But I think there’s still going to be a huge amount of value in learning how to manage and use these systems for the foreseeable future.

AI has an incredible ceiling, it can pretty much make anything, but that means nothing if the person behind the wheel is driving in the wrong direction. Just because you won’t be the one writing the code doesn’t mean the underlying principles are useless. Having a deep understanding of what actually makes software good will still be a highly in demand skill.

If you have heard of OpenClaw you’re already a top 1% AI user. Don’t fall behind. Learn these new tools. Play with them, see what they can do, even if they aren’t a perfect fit for your personal use case. To close out, here’s a quick message from Batman…

Thanks so much for reading this edition of Launch Letter. I know it's been a while (rest in peace the weekly schedule haha), hoping to start posting on here again if there's any interest from you guys to read it so let me know.

Kirby (and Batman) out.