I have seen people share prompts to change their OpenClaw soul files. A prompt can update the default configuration and make the software better. That is interesting, but it also raises a different question for me.
Do we need OpenClaw at all? Or can we use Linux itself as the base?
Maybe the operating system is enough
Linux already has a lot of what agent software needs. With systemd
in user mode, services can be started, restarted, and kept running. It also
supports timers, so we do not need a different cron runner just to schedule
work.
Then there are already files, logs, scripts, permissions, networking, and a shell. That is a lot of infrastructure before we add any special agent platform on top.
A coding agent plus skills might be enough
The stack may be simpler than people think: a Linux system, a coding agent, and a set of skills and scripts. That may already be enough to build a lot of useful software.
Instead of starting with a whole new product layer, we could start with the operating system and let the agent become the glue.
Do we need Slack-shaped software?
The same question applies to chat interfaces. Why do we need Slack? One big reason is mobile access and the idea of being connected everywhere. But an HTML-based chat can also show images. Terminals can do this as well.
When I use pi, it can show the images it reads, the links can be clicked, and the text streams in live. Slack can hide that a session starts in the background, but pi can do this too with some extensions. The problem with Slack is that I can't stop what is happening. With pi I just tap Escape and it stops.
This is also about running software on your own computer. When I run the agent on my own machine, I can log into a website and let it control the browser, or let a tool connect to Gmail and read an email. What I still wonder about is how to make this useful for more people when the software runs on a VPS. In a team situation, maybe everyone should have their own agent and software that is logged into the services that person uses.
Malleable software
This is what I like about malleable software. We do not have to accept the first shape a tool comes in. We can extend it, rearrange it, and adapt it to the way we actually work.
Pi is a good example. We can build extensions for pi by using pi. The tool can help reshape itself. That makes it feel much softer than software that forces you into one fixed structure.
Fewer layers, more room to change
I keep coming back to the same thought: maybe we do not need so many extra layers. The operating system already gives us a strong base. The agent can work with that base directly. Skills can capture repeated patterns. Small scripts can handle the parts that need automation.
That sounds better to me than adding another platform that freezes the work too early.