Why I Scrapped OpenClaw

OpenClaw was meant to be a light-touch way of automating bits of work for me.

That was the whole point. It wasn’t supposed to be a project in itself. It wasn’t meant to become another thing that needed managing. It was supposed to sit quietly in the background and make a few useful things easier.

The problem was WhatsApp.

The way OpenClaw integrated with it meant it was watching WhatsApp all the time. That sounds fine until you live with it for a bit. If a message came in, WhatsApp had effectively already been seen by the time I looked at it on another device, so I wouldn’t get the normal notification flow I rely on.

That got annoying quickly.

It also meant the integration was doing too much. I didn’t want a setup where the useful bit of automation came bundled with a constant sense that my messages had been touched before I had seen them properly. That might sound small, but it changes how relaxed the whole thing feels.

There wasn’t a reliable workaround that kept the setup simple.

The better answer would have been to give OpenClaw its own dedicated number, or move the whole thing to something like Discord or Telegram. Those would have been cleaner from a technical point of view. But they would also have made the thing feel less light-touch, which was the part I cared about most.

If a workflow starts feeling like a chore, it’s already lost something important.

So instead of forcing it, I scrapped it.

So I pulled the workflows back into this repo as agent files and skills. That keeps the useful bits close to the site and makes them easier to reuse without dragging the whole OpenClaw setup along with them.

That also means the writing process for this blog post is now using the same repo-local setup. Which feels neater, honestly. The blog, the workflows, and the instructions for how I want things done all live together now.

That seems more sensible than having the system half-live in a messaging app and half-live in the repo.

There’s also a broader reason for stepping away from it now.

Codex, Claude, and the other tools in this space are getting better at handling the sort of automation I was trying to make OpenClaw do. That doesn’t mean I’m done with automation. It just means the shape of it is changing a bit.

So I’m going to keep exploring that side of things, just in a different way.

For now, though, OpenClaw is done. It did enough to prove the point, and it also made the limit pretty clear. If the integration layer becomes the thing that gets in the way, the simplest answer is usually to remove it and keep the part that actually helped.

That’s what I’ve done here.