The Beginning of the Agent's Final Form

We have reached the beginning of the agent's final form, with three core components

agentic ai coding OpenClaw

Previous Assumption

Prompt science is overblown. Because prompt science was based on prompting 'Chat', and AI can be prompted by AI.

New Assumption

The prompt is the Markdown 'MD' files. Which will be grown and honed by individuals and agents - this is a core component of the agent. We're not learning to prompt better, but we will develop our MD files, like an influencer pours over building their social media profile. The agent is self-aware and self-improving.

What an Agent Actually is

1. Markdown files of user preferences, history i.e. their personality and are used as the metaprompt for the LLM 2. An LLM model whose job is to command actions on your computer 3. The Unix shell on hardware with cron jobs programmed in, and which the LLM communicates with

The crazy part? All three of these are modular - they can be swapped out, interchanged for like-components, substituted.