Ruby Rogues Autogenetic AI Agents and the Future of Ruby Development - RUBY 682
9 snips
Dec 24, 2025 
Valentino Stoll, a software engineer building agentic AI systems and the Ruby gem agentic, explores autogenetic self-assembling agents. He discusses plan-and-execute workflows, agent reuse and capabilities, and how local powerful models blur development and production boundaries. The conversation covers protocols, guardrails, self-healing agents, and what developers should prioritize as AI changes tooling.
AI Snips
Chapters
Transcript
Episode notes
Agents That Self Assemble From A Task
- Autogenetic agents let an LLM design and assemble specialized agents on demand instead of developers predefining each role.
- Valentino's agentic gem produced named agents from a task prompt and saved them as reusable artifacts for later use.
Limit Capabilities To Improve Determinism
- Register a bounded set of capabilities so agents use known, testable tools instead of inventing unreliable ones.
- Valentino implemented a capability system (search web, read file, generate PDF) and limited available capabilities to improve determinism.
Experiment Where The System Built Its Own Dev Team
- Valentino ran a self-assembly experiment and the system built a Ruby coding agent plus PM and QA agents to scaffold the project.
- He reused some of the generated agents in his own projects and ran models locally (StarCoder, Olama) to avoid cloud cost.