saved
What If We Took Message-Passing Seriously?
Hraness wrote this summary from a saved copy of the source. Quotations are taken word for word from the source.
gist
Scott Werner reframes AI agents as Smalltalk-style prompt objects: self-contained environments that receive natural-language messages, interpret meaning at runtime through semantic late binding, and may modify themselves. Drawing on Ruby and Alan Kay, he argues agent primes reliability and guardrails while prompt object primes composition and message protocols. His Ruby gem prompt_objects starts with receive, think, self-modify, and create, then bootstraps capabilities from the environment. He claims LLMs may restore living, reshapeable Smalltalk-like places where interfaces are negotiated in prose, and ships a rough open-source gem to explore that detour.
ideas
- Prompt objects, not task agents. Treat an agent as a self-contained computing environment that receives messages and interprets them, rather than an LLM-with-tools pointed at a job.
- Message passing is natural language. Communication is not structured function calls; the receiver decides what a message means at runtime (semantic late binding).
- Naming reshapes design. “Agent” primes autonomy, tasks, and guardrails; “prompt object” primes composition, interfaces, inheritance, and message protocols.
- Self-modification is the default. A minimal prompt object can receive, think, modify itself, and create others, then add capabilities by querying its environment.
- LLMs revive living Smalltalk places. Soft boundaries and negotiable interfaces let objects describe capabilities in prose and bootstrap competence while the system runs.
quotes
“Objects as little computers. Messages passed between them, interpreted by the receiver.”
“Let the receiver decide what the message means.”
““Agent” primes you toward autonomy, tasks, guardrails, going rogue.”
“Self-modification as the default state.”