S12 Favorite - Making AI Deterministic for Developers and their Agents, with Patrick Vuong of Moderne
Summary
Patrick Wong, Director of Product at Modern, discusses how the company provides deterministic tooling for AI agents to ensure accuracy and efficiency. He explains Modern’s shift from tech debt remediation to building a "deterministic harness" using lossless semantic trees and programmable recipes. This approach helps agents navigate complex codebases, reduce token costs, and integrate safely into the software development lifecycle.
Topics discussed
Sponsors: .tech domains, Unblocked, Mesmo, and BrainGrid
Guest Intro: Patrick Wong's background at Microsoft and Modern
Modern's evolution from tech debt cleanup to AI agent tooling
Why agents need deterministic tooling for accuracy and efficiency
Lossless Semantic Trees and deterministic recipes for code
Modern as a deterministic harness in the agent-driven SDLC
Target customers: From code readiness to autonomous factories
Mid-roll sponsors: Unblocked, .tech, BrainGrid, and Mesmo
Product suite: TriGrap, PreThink, and ChangeLog via MCP
Future roadmap: Multi-repo search, autonomous factories, and governance
Market fit: Serving startups and enterprises with open source roots
Patrick's transition from Microsoft to the startup world
Inspirations: Kobe Bryant's obsession with craft and detail
Advice: Creating a category story rather than just shipping features
Outro: Summary of Modern's deterministic approach and links
Sponsor: Capital One's multi-agentic AI Chat Concierge
Listen ad-free on Castria