Active Agent with Justin Bowen
Notes de l’épisode
Seventeen-year Ruby veteran Justin Bowen joins hosts Valentino Stoll and Joe Leo to unveil Active Agent—a Rails-native framework that treats every agent like a controller and every prompt like a view, letting you weave LLMs, vector search, and business logic straight into MVC.
The crew also digs into the real-world mechanics of shipping AI: defining ground-truth datasets, replay-ready evaluation harnesses, and tight retry logic that keeps hallucinations out of production. You’ll hear a candid take on the current hype cycle (and its parallels to crypto), the challenges of long-term gem maintenance, and fresh ways to keep open-source sustainable—think GitHub Sponsors, corporate grants, and pro-tier gems.
What you’ll hear
- Active Agent 101 – agents as abstract controllers, templated prompts as views
- Testing in the wild – fingerprints, VCR cassettes & CI pipelines for non-deterministic code
- Context is king – why ground truth matters when counting cows or parsing legal docs
- OSS meets ROI – balancing passion projects with sustainable monetisation
- Rails vs. Python/Next.js – reclaiming the one-person startup stack
- Community fuel – Discords, hackathons, and the push for academic & corporate sponsorship
Mentioned In The Show:
- Active Agent (GitHub) – Justin’s Rails-native, agent-oriented framework for building AI features.
- Vercel AI SDK – TypeScript toolkit whose generative-UI ideas helped inspire Active Agent.
- Maestra.ai – YC W24 startup offering AI transcription, dubbing, and hosted agent runtimes.
- Matz's 2025 Ruby Kaigi AI Keynote
- ONNX Runtime Ruby – Gem that runs ONNX models (CPU/GPU) from Ruby.
- PGVector gem – Ruby bindings for PostgreSQL’s pgvector extension (embeddings storage).
- Neighbor gem – k-NN / ANN vector search for Rails & Postgres—pairs nicely with PGVector.
- Hugging Face JS – Run models in the browser with WebGPU and ONNX
- Hugging Face Spaces – No-config platform for hosting ML demos; handy for sharing agent prototypes.
- LangSmith (LangChain) – Evaluation & observability service discussed as a monetization model.
- CrewAI (GitHub) – Python framework for orchestrating multi-agent “crews”; Joe’s current go-to.
- Honeybadger – Rails-first error-monitoring SaaS—an inspiration for future Active Agent services.
- Rising Impact – A Netflix anime special about a third-grader's journey to be the world's best golfer.
- Osmo AI – Google-born startup using AI to digitise smell—cited in the show’s “AI hype” chat.
- Ruby AI Builders Discord – Public Discord community for Rubyists building AI apps.