Episode

Making AI Deterministic for Developers and their Agents, with Patrick Vuong of Moderne

Podcast
Code Story: Insights from Startup Tech Leaders
Published
May 7, 2026
Duration seconds
1471
Processing state
processed
Canonical source
https://codestory.co/podcast/making-ai-deterministic-for-developers-and-their-agents-with-patrick-vuong-of-moderne/
Audio
https://pdst.fm/e/pscrb.fm/rss/p/audio4.redcircle.com/episodes/68d89dbd-247b-4679-b624-906c64ece1c7/stream.mp3
JSON
/v1/public/podcasts/code-story/episodes/making-ai-deterministic-for-developers-and-their-agents-with-patrick-vuong-of-moderne
Markdown
/podcast/code-story/making-ai-deterministic-for-developers-and-their-agents-with-patrick-vuong-of-moderne.md

Actions

  • POST https://stenobird.com/v1/public/podcasts/code-story/episodes/making-ai-deterministic-for-developers-and-their-agents-with-patrick-vuong-of-moderne/transcription-requests
    Idempotently request low-priority transcript generation for this episode.
  • GET https://stenobird.com/podcast/code-story/making-ai-deterministic-for-developers-and-their-agents-with-patrick-vuong-of-moderne.md
    Read the agent-friendly Markdown representation of this episode resource.

Summary

AI agents lack the structure and context required to safely operate on enterprise codebases at scale. Moderne provides a deterministic 'harness' using semantic models to ensure agent-driven changes are predictable, auditable, and accurate.

Topics

  • AI Agents
  • Software Development Life Cycle
  • Deterministic Computing
  • Semantic Models
  • Agentic Workflows
  • Code Automation
  • Software Engineering
  • Product Management

Highlights

  • Main idea: Moving from probabilistic inference to deterministic execution via semantic models
  • Practical takeaway: Use 'recipes' and symbol-aware search to reduce token consumption and increase agent accuracy
  • Failure mode: Relying on pure code generation without a structured harness leads to broken features and high correction loops
  • Main idea: The SDLC must evolve from human-centric workflows to agent-orchestrated 'dark factories'
  • Practical takeaway: Focus on building a category-defining product story rather than just shipping individual AI features

Chapters

  1. 3:30 Introduction to Patrick Vuong: Patrick discusses his background and transition from Microsoft to the startup world.
  2. 5:50 Defining Moderne: An overview of Moderne's mission to enable developers to operate software at the speed of agents.
  3. 8:10 The Agentic Landscape: Exploring the shift toward agent-driven development and the need for orchestration.
  4. 13:00 Deterministic Recipes: How semantic models and reusable recipes allow agents to perform complex, large-scale migrations.
  5. 15:20 The Deterministic Harness: How Moderne acts as a governance layer to ensure agent work is auditable and lands correctly.
  6. 19:50 Improving Agent Search and Context: Technical details on Trigrep and reducing token usage through precise, symbol-aware code search.
  7. 29:30 Career Transitions and Product Strategy: Advice on transitioning from big tech to startups and building a cohesive product category.