AI for Founders with Ryan Estes

The episode contains some valuable practical points about AI architecture, modular design, and the dangers of 'vibe coding,' but suffers from significant filler, tangential discussions, and repetition. The core insights about guardrails, not locking into a single model, and the difference between MVP and production-ready products are solid but not densely packed. Multiple sponsor reads and off-topic commentary (World Cup celebrations, Omega-3s) dilute the substance significantly.

“Speed without architecture doesn't scale, it just delays the collapse”

“you cannot save the costs, uh, and the knowledge of a real Architect, you can um, save the costs and the time of the low level code developers”

The discussion treads familiar ground in AI/startup discourse: MVP vs. production code, technical debt, avoiding vendor lock-in, using multiple models for different tasks. While the framing around 'vibe coding' is catchy, the underlying concepts are not novel. The guest restates conventional wisdom about architecture and guardrails without offering genuinely fresh frameworks or contrarian takes.

“vibe coding your MVP feels like progress, but it's actually setting a time bomb under your business”

“you can use the right model for the right tasks”

Mike Vites is a legitimate practitioner running a custom software/AI services firm (Saturnia Design) with real client work and products in development. He has technical credibility (computer scientist, architect background) and is building deployed systems in production. However, he is not a massive-scale operator or household name in the space, and the episode reveals he's primarily a service agency founder rather than a product founder at unicorn scale.

“founder of Saturnia Design and Saturnia Labs, a company that builds custom AI power software and Agentix systems”

“we are using Fable, but we are not saying that, okay, Fable is the most intelligent model”

The episode includes some concrete examples (chat support system with 700 questions answered, 83% automated; 40 parallel agents for research; customer support use case), but lacks specifics in many other areas. Numbers are provided sparingly, and many claims remain abstract (e.g., 'we're building an agentic system' without technical detail, 'upcoming products under NDA'). The examples given are helpful but not dense enough.

“we rolled it out I don't know three weeks ago and it already solved 700 questions and the uh, 83% was automatedly answered”

“we've been using more than 40 agents parallel to do the research”

The host (Ryan Estes) asks generally reasonable questions but rarely presses back, challenges, or follows up with real skepticism. When Mike makes claims (e.g., about model switching, automated CI/CD), Ryan accepts them without pushing for implementation details, trade-offs, or potential downsides. The conversation meanders - discussing World Cup celebrations, tangential product ideas, and sponsor reads - rather than drilling into substantive disagreements or edge cases.

“Yeah, absolutely”

“Cool. And what kind of projects are you really looking forward to?”