AI agents · LinkedIn adaptation
What changes when you grow an AI agent?
An OpenClaw experiment on how agents can develop new capabilities as people use them.
In his LinkedIn post, Ravi describes how he used to think about agents in terms of jobs. A support agent, an underwriting agent, or a sales agent would have a defined task, a prompt, and the tools to carry it out.
What changed with OpenClaw
One agent became useful across several kinds of work. It tracked freelance time by client, carried out research, and prepared documents. When it lacked a capability, Ravi could ask it to build one or look at an existing framework. The experience shifted his attention to what an agent could learn to do over time.
What happens to the tools underneath?
Ravi compares the shift to the rise of general-purpose computing. If agents become the place where people coordinate work, business applications may play a bigger role as reliable systems of record with standard interfaces. It is a hypothesis about where software could go, and it raises an immediate design question: what should an agent be able to read, change, or hand off?
A small experiment to try
Pick one recurring, reversible task, such as turning your own meeting notes into a checklist. Keep the input, the output, and the review criteria. Try it again after a week. Add a tool or permission only when you can name the problem it solves. Use the worksheet below to write down what counts as success before the first run.