Building Private AI: Ivan Lee on AI Agents, Open-Weight Models, and Data Ownership

Damien Filiatrault speaks with Datasaur founder Ivan Lee about the growing importance of private AI and why businesses are reconsidering their reliance on external AI providers. They discuss how organizations can build secure AI systems using open-weight models, automate complex workflows with AI agents, and balance efficiency with human oversight. Ivan also explains the technical decisions behind custom AI development and why owning AI models, data, and infrastructure could become a long-term competitive advantage.

Damien Filiatrault talks with Justin Gelinas, Chief Business Officer at TAHO Labs, about the hidden inefficiencies driving up the cost and energy demands of AI infrastructure. They explore how better workload distribution, improved hardware utilization, and more efficient compute could help companies get more from existing resources while unlocking new opportunities for AI and software development.

Eric Ries joins Damian Filiatrault to discuss the evolution of startup thinking from The Lean Startup to his new book, Incorruptible. The conversation explores how founders can test ideas ethically, why growth is not the same as validated learning, and how companies often lose trust when short-term financial pressure takes over. Eric also explains why governance, mission, and ownership structure matter much earlier than most founders think—and how builders can create organizations that are not only successful, but worth protecting.

Damien Filiatrault talks with Micah Johnson, co-founder of Biggest Goal, about what AI agents actually are, how they differ from basic chatbots, and where they can create real value for businesses today. The episode covers practical use cases like data analysis, internal knowledge retrieval through RAG, and event-driven automations, along with why tools like N8N and Raggy have made agentic workflows much more accessible. It also looks at why so many AI projects fall short: not because the technology is weak, but because companies often adopt it without the systems, training, and strategy needed to make it work.

In this episode, host Damien Filiatrault talks with Kevin Williams, founder of Ascend AI, about what “agentic commerce” actually means as shopping and discovery start shifting into ChatGPT- and Gemini-style interfaces. They break down the path from today’s LLM-assisted recommendations to “zero-click” transactions that happen entirely inside an AI chat, without users ever visiting a traditional checkout flow. Kevin explains why brands are seeing meaningful declines in organic traffic, why being discoverable now requires a mix of AEO (answer engine optimization) and GEO (generative engine optimization), and how those strategies extend beyond your website into reviews, sentiment, and the broader web. They also dig into the messy reality of emerging standards (product feeds, schema, and competing protocols) and what builders may actually need to implement to make products legible and purchasable by agents. If you’re building for e-commerce, marketing, or the next web experience, this episode offers a practical look at what’s changing, what’s still unsettled, and how to prepare before “zero-click” becomes the default.

In this episode, host Damien Filiatrault talks with Brit Myers, VP of Engineering at System Initiative, about rethinking infrastructure automation for a world of legacy systems, real-time fixes, and AI collaboration. They explore why traditional Terraform and GitOps assumptions break down in practice, how System Initiative’s “digital twin” model keeps intent, state, and change tightly coupled, and why simulations matter before anything touches production. Brit explains how programmable TypeScript functions and change sets make it safer to work with AI agents—letting them propose and iterate on infrastructure changes without handing over full control. If you’re navigating cloud complexity, drift, or experimenting with AI in DevOps, this episode offers a grounded look at how teams can move faster with more confidence and a lot less “merge and pray.”

In this episode, host Damien Filiatrault sits down with Coder CEO Rob Whiteley to explore how AI agents are changing the way software gets built, from cloud-based dev environments to long-running “software intern” workflows. They unpack how Coder provisions secure, centralized workspaces for both humans and agents, why Anthropic’s Claude Code became so effective once it could read its own Terraform-defined context, and how MCP-powered toolbelts turn a basic agent into a capable teammate. Rob breaks down why code completion alone is yesterday’s story, how senior engineers orchestrate multiple agents like a conductor, and why startups often stick to cursor-style assistance while enterprises layer in Bedrock, governance, and stricter controls. If you care about developer productivity, you’ll come away with a realistic view of where agents help today, what it takes to trust them with real projects, and why “English as the new programming language” opens the door for many more people to build software.

In this episode, host Damien Filiatrault talks with Josh Clark and Veronika Kindred of Big Medium about “Sentient Design”—interfaces that adapt to context and intent without ceding user control. You’ll hear a cross-generational take on rethinking UX best practices, from bespoke UIs that assemble trusted design-system components on the fly to the NPC pattern where AI participates alongside humans. They dig into defensive design—guardrails, reversibility, explainability, and graceful degradation—plus a practical split: let LLMs interpret intent, not act as the source of truth. If you’re building with AI, expect concrete patterns for smarter, safer products that keep suggestions deferential and users firmly in the driver’s seat.

Founder Iddo Gino tells host Damien Filiatrault how a GitHub repo called “Awesome APIs” became RapidAPI, a global marketplace later acquired by Nokia, and what sky-high valuations actually change credibility and hiring more than outcomes. He argues that instead of chasing new standards like MCP, teams should fix the REST/GraphQL APIs they already have with accurate specs and living docs. Gino then explains Datawizz, his new platform that routes requests to tiny, task-specific models and falls back to large LLMs only when needed, often cutting costs by 85–95% while improving latency and reliability. He details an OpenAI-compatible router, when volume justifies specialization, and how edge/on-device options, Cloudflare AI Workers, iOS’s built-in models with adapters, and Chrome’s Gemini Nano, unlock faster, cheaper, and more private inference.

Developer and founder Killian Valkhof tells host Damien Filiatrault how a homemade tool that made his work 60 percent faster became Polypane, a dedicated browser for web professionals. He outlines Polypane’s core features—side-by-side responsive views, built-in accessibility and performance audits, and instant social-media previews—and shares the tactics that keep his one-person, bootstrapped SaaS both profitable and user-driven. Valkhof explains his open feedback channels, no-credit-card trial, and ROI calculator for team budgets, describes spin-off utilities like Superposition and FixA11y, and reflects on why he is holding off on AI features in favor of polished fundamentals.