OpenClaw, Post-Hype Report | ✉️ #89

Illustrated image featuring "MKDEV DISPATCH #89" with text "Openclaw, Post-Hype Report" on an orange and gray background with paper airplanes. On the right, a smiling person holds a small furry animal with a paper airplane above them.

Last updated: 14/03/26 | Published: 14/03/26

Hey! 👋

Of course we tried OpenClaw. In my case, I have it running right behind my back, on an old ThinkPad Laptop.

I was hesitant to try it at first, because it was hard to see how it’s different from just using Claude Code - which I already use way beyond writing code. But the convenience of accessing it via Telegram convinced me to try.

Unfortunately, I came to a conclusion that OpenClaw is not exactly the tool that solves much for my professional life. Perhaps, it’s because it was built to be a personal assistant - and yet, I failed to come up with any personal use case for it. Checking weather? Boring. Checking news? Perplexity does it for me. Kicking of research, writing mini apps, reading my emails? Nothing what Claude or ChatGPT can’t do. It almost feels like a trick: lots of people got convinced, that in the personal assistant space OpenClaw is doing much more, than what Claude and ChatGPT already do. Perhaps, I am a wrong kind of user - I already maximise my AI usage in my personal life, and I could not see how OpenClaw gives me more that what I already have.

But professional use cases are different. This is where OpenClaw has a lot of appeal. You are not kicking off the agents yourself anymore - you can setup a system, where the agents are working on their own. Posting Instagram posts. Picking up tasks from Linear or Jira. Filling in the forms on app directory websites. Logging all of your conversations and analysing them for proper reflection (alright, this one is a bit more personal than professional).

And yet even for professional use case, OpenClaw is not there yet. It burns tokens like a maniac. It mixes things up, because it combines sessions with a single chat box in your favorite messenger. And it requires more tooling. It needs work - as any open source tool. But it can also be made uniquely yours - unlike any none-opensource AI assistant.

So I’ve been working on a new level of maximising my AI usage, this time by building new tools and systems around OpenClaw. OpenClaw is a decent AI execution engine - it supports any model, any subscription, and it has this nice session-based architecture. Around this engine, we can build a lot of custom tooling that brings us even further. OpenClaw is the biggest AI release of the year, and it has zero AI in it - it’s not a model, it’s just an evolution of a tooling around models. And it this openness and extensibility that makes this evolution feel like a revolution.

While unimpressive on a technical level, OpenClaw certainly is shifting the paradigm. Almost every company since the release of OpenClaw introduced new features around their agents, that let the agents work on their own. This is the next stage we just reached - we are not using AI as a tool, we equip AI with tools, and then we work alongside AI agents, almost the same as we would work with humans. At least this is what we are building now at mkdev.

The 90th mkdev dispatch will arrive on Friday, March 27th. See you next time!