Ideas for an Agent-Oriented Programming Language
3 months ago
- #Programming Languages
- #AI Agents
- #LLMs
- Agent-oriented programming language 'Markov' designed for LLM strengths and human readability.
- Emphasis on human-readable and editable code for transparency and interpretability.
- Use of sum types and static analysis for fast compiler feedback to agents.
- Compiler errors optimized as prompts for LLMs with suggested fixes.
- Token-optimized syntax to reduce costs and improve LLM understanding.
- Interoperability with existing ecosystems considered a non-goal due to agent capabilities.
- Functional programming idioms may enhance token efficiency.
- Agents can propagate changes through codebases, reducing boilerplate work.
- Potential for embedded DSLs within strongly typed languages.
- Future vision includes agents independently building systems end-to-end.