Natural-language context and prompts become a programming medium when LLMs can interpret them as instructions, constraints, examples, and executable intent.

Summary

In Software 1.0, humans write explicit code. In Software 2.0, behavior is learned in neural network weights. In Software 3.0, a human supplies context, instructions, examples, files, and constraints to an LLM that interprets that context and acts in the digital environment.

Why It Matters

The unit of leverage shifts from exact code to effective context. A good instruction to an agent can replace a long shell script or manual setup guide because the agent can inspect the environment, adapt, run commands, and debug.

For this wiki, AGENTS.md, notes/index.md, and log.md are Software 3.0 artifacts: they program future LLM behavior through structured context.

Sources

Open Questions

  • What parts of this repo are best thought of as Software 3.0 instructions?
  • How should prompts, specs, and agent instructions be versioned?