OpenClaw Episode 3: The Doubt
The doubt phase: questioning whether OpenClaw was just a harder way to use ChatGPT, and the turning point when planning the personal site revamp revealed the real difference.
After the initial excitement wore off, I hit a wall.
I had been using OpenClaw daily, talking to agents, testing setups, and trying different models. But one evening, sitting on the couch with my phone, I caught myself asking a question I could not shake: What is the actual difference between this and ChatGPT?
It is just a chat. A conversation. ChatGPT can also be configured for agent behaviour. It remembers context. It drafts things. It answers questions. What was I doing here that I could not do there?
I felt dumb. I had spent days setting up infrastructure, tweaking prompts, connecting Telegram, managing self-hosted infrastructure. For what? To get responses that felt similar to what I could already get from a polished consumer product.
The excitement that carried me through the setup phase suddenly looked like noise. I started to think maybe this whole experiment was just a more complicated way to do something that already worked fine.
I almost stopped.
But I had one thing left to try. I had been thinking about revamping my personal site. It was long overdue. I decided to ask the agent to help me plan it. Not just generate content, but actually plan the project. Architecture, design phases, tech stack decisions, build steps.
What came back was not a generic template. A plan shaped around my actual requirements, constraints, and preferences. That was the first time I felt the difference. Not because the agent was smarter than ChatGPT. But because it was mine. It knew my context. It had been part of my setup. It was connected to tools I could actually use.
The plan was not just text in a chat window. It was a starting point for execution.
A few days later, I came back and asked the agent to start building it. I wanted the relevant agents to pick up tasks from the plan and execute them. That was when a new idea clicked: if agents were going to do real work, I needed a way to track it. Not a full project management suite. Just something simple. A Kanban board. Something that showed what was queued, what was in progress, and what was done.
That thought, I need a Kanban board for my AI agents, was the moment the doubt disappeared. I was no longer comparing OpenClaw to ChatGPT. I was building a system.