378% more clicks. We published four blog posts that whole month. Four. Almost everything came from fixing pages that were already ranking and just quietly getting ignored.
All I wanted was an AI agent to check a few things every morning and send me a summary. Simple right?
Three hours later I'm watching Docker tutorials on YouTube, editing config files I don't understand, and Googling error messages that make no sense to me. I'm a marketer, not a DevOps engineer.
Three months ago we launched a free plan for BetterClaw (we're an AI agent platform, launching here on the 23rd). Every feature included. No trial countdown. No credit card. You bring your own LLM API key, so we never touch your model bill and your conversations never route through anything we can read.
I genuinely thought the response would be "cool, thanks."
The first comment on our launch post was "what's the catch." Not a question. A statement. Like I was a guy in a trench coat handing out free candy.
I replied that there's no catch, whatever the free plan costs us nothing significant because users bring their own keys. The response: "That's exactly what someone with a catch would say."
But after building BetterClaw, I think the more interesting question is:
How autonomous should an AI agent actually be?
If an agent can send emails, update CRMs, trigger payments, or modify production systems, giving it unlimited permissions feels convenient... until something goes wrong.
I've been thinking about this a lot lately. We give AI agents access to our email, our CRM, our calendars, our files... and then we go to sleep.
Most of us have no idea what the agent is actually doing at 3am. Is it running normally? Did it enter a loop? Did it send something it shouldn't have? We just wake up and check the output.
Most no-code agent platforms bundle a model and charge you a markup on every token you use. It's an easy way to make more margin and it means less setup for the user. We didn't do that with BetterClaw.
We went BYOK, so you bring your own API key and use whatever model you want. Honestly it made the build harder. More provider quirks to handle, more support questions about setting up keys, slower onboarding overall.
But almost everyone we spoke to while building already had a model they liked or wanted to test two or three before settling on one. Forcing them into whatever we picked just didn't sit right.
Under the hood it runs on Rust (ZeroClaw), mainly because we wanted agents to handle longer background tasks without choking. Might write that up separately sometime.
