AI Agents Go Mainstream
(And Why the UI Still Sucks)
TL;DR
OpenClaw (formerly Clawdbot) just became one of the fastest-growing GitHub repos ever — 100,000 stars in two months. Mac Minis are selling out as enthusiasts set up dedicated agent machines. But here's the reality from someone actually running one in production: the capabilities are a 10/10. The UI? Maybe a 2. This is still early adopter territory, and that's exactly why it matters.
What Up Y'all?
If you've been anywhere near tech Twitter this week, you've seen the headlines. OpenClaw — the open-source AI agent that runs locally and connects to your WhatsApp, email, calendar, and pretty much everything else — just hit 100,000 GitHub stars. Retailers are reporting Mac Mini shortages. And oh yeah, someone launched Moltbook, a social network exclusively for AI agents where 1.5 million bots are chatting while humans can only watch.
It's a lot.
But here's what nobody's talking about: I've been running one of these agents for months now. Not on a shiny new Mac Mini — on an HP all-in-one Windows PC. And my take is probably going to annoy both the hype crowd and the skeptics.
The capabilities? Utterly amazing.
The product? The UI kind of sucks.
The Real Practitioner's Take
I run Fitted Automation, a custom dev shop focused on AI implementations. So when I say I'm running an AI agent in production, I mean it's actually doing work. Not demo work. Real work.
Here's what my agent handles daily:
Every morning at 6:25 AM:
- • Reads my meeting notes from the last week
- • Checks my Gmail and flags urgent emails by client
- • Pulls my calendar for the day
- • Validates deployment status across GitHub, Railway, and six Replit projects
- • Updates my to-do files per client
- • Sends me a WhatsApp briefing before I've had coffee
Ten minutes later:
- • Syncs my Podio CRM
- • Enriches new leads with web research
- • Scores them on "warmth" (fit for my ideal customer)
- • Drafts personalized outreach
- • Asks for my approval before sending anything
Throughout the day:
- • Drafts emails I can send with one click
- • Creates GitHub issues when clients report bugs
- • Monitors for new leads every two hours
- • Stays quiet when there's nothing actionable
This isn't hypothetical. This is running. On Windows. Right now.
Capabilities 10/10. UI 2/10.
Here's my hot take, and I'll stake my reputation on it:
10
UX Score
The UX — what the agent actually does — is transformative. I'm getting 13+ hours a week back. My morning briefing alone saves me an hour of context-switching across five different apps.
2
UI Score
The UI — how you set it up and interact with it — is rough. We're talking YAML configuration files. PowerShell scripts. Debugging API integrations. Learning the difference between a "heartbeat" and a "cron job."
This is not a plug-and-play product. It's a capability waiting for a product.
And honestly? That's fine. That's where we are. But let's stop pretending this is ready for your accountant.
The Windows Angle Nobody's Writing About
Everyone's buying Mac Minis. The tech press is writing about San Francisco retailers selling out of them.
I'm over here on an HP all-in-one. Windows 11. And it works great.
Why? Because OpenClaw runs on Node.js. The agent doesn't care about your hardware religion. It cares about whether you can configure a cron expression and troubleshoot a failed API call.
The Mac Mini shortage isn't a hardware story. It's a demographic story.
The early adopters who set up AI agents happen to skew Mac. That's it. Windows works fine. Linux probably works better.
If you're waiting for the "right hardware," you're waiting for the wrong thing.
Who Should Try This (And Who Should Wait)
Let's be real about who this is for right now.
Try it if:
- • You're technical enough to edit config files without panicking
- • You actually have repetitive workflows worth automating
- • You're willing to iterate and debug for a few weeks
- • You understand this is a capability, not a finished product
Wait if:
- • You want something that "just works" out of the box
- • Your business doesn't have clear, repetitive processes yet
- • You'd rather pay for a polished SaaS than tinker
- • The phrase "YAML configuration" makes you nervous
There's no shame in waiting. The UI will get better. Twelve months from now, there will be products built on top of this that don't require any configuration. But right now, the value is captured by people willing to push through the rough edges.
The Frontier Is UX, Not UI
Here's the bigger point:
We're at a moment where AI capabilities have massively outpaced AI interfaces. The models can do incredible things. The agents can orchestrate complex workflows. But the way humans interact with them is still stuck in developer-tool territory.
This is the frontier. Not the models. Not the training data. The interface.
The companies that figure out how to wrap these capabilities in genuinely usable products — not just chatbots, but actual workflow tools — are going to win the next phase.
Right now, early adopters like me are doing the integration work manually. We're the beta testers. And yeah, the UI sucks. But the UX? The actual experience of having an AI employee that works 24/7?
That's worth a few YAML files.
Key Takeaways
OpenClaw hitting 100K stars is real — but it's still early adopter territory
Capabilities are incredible; UI/setup experience is rough
You don't need a Mac Mini — Windows works fine
This is a capability waiting for products, not a product yet
The real frontier is interface design, not model capability
If you're willing to tinker, the ROI is already there
FAQ
Are AI agents worth it for small business?
Yes — if you have clear, repetitive workflows and someone technical enough to set them up. The ROI is real (I'm saving 13+ hours/week), but the setup cost is significant right now.
Do I need a Mac Mini to run an AI agent?
No. OpenClaw runs on Node.js, which works on Windows, Mac, or Linux. The Mac Mini shortage is a demographic quirk, not a technical requirement.
Is OpenClaw a security risk?
It requires broad system access, which creates prompt injection risks. Use it on a dedicated machine or sandbox if you're handling sensitive data. The security model is still maturing.
How long does it take to set up an AI agent?
Expect a few weeks of iteration to get workflows dialed in. The initial install is quick, but configuring it for your specific needs takes time.
Should I wait for better AI agent products?
If you're not technical, yes. Better products are coming. If you're willing to tinker, the value is available now.
About the Author
Ryan Duffy is the founder of Fitted Automation, an AI automation consultancy helping small businesses implement intelligent workflows. He's been running AI agents in production since before it was cool (and before the UI got any better).
February 2026