Futuristic Math Will Power Agentic Swarm Coding: Carina Hong is Building it at Axiom Math

Futuristic Math Will Power Agentic Swarm Coding: Carina Hong is Building it at Axiom Math

Oct 9 · 20 min

People

  • Raphaëlle d’Ornano Host
  • Carina Hong Guest

Watch on

Suggest an edit

Carina Hong, CEO of Axiom Math, joins Raphaëlle d’Ornano, CEO & Founder of @DecodingDiscontinuity on stage to explain why the stack that proves mathematical theorems is the same stack needed to verify AI-written code. She covers Axiom’s new world record on bounded gaps between primes, the infrastructure bottleneck facing swarms of autonomous coding agents, and how intent-based programming turns human intent into theorems an AI prover can check. 00:00 - Verifying the Value of AI: Mathematical Discovery Meets Verification 00:53 - AI Broke a Math Record Researchers Have Chased Since 2013 03:28 - Why the Stack That Proves Math Theorems Also Verifies Code 05:07 - Why a Million Autonomous Agents Need Formal Guarantees to Build on Each Other 06:12 - Why Agent Swarms Hit a Software Infrastructure Bottleneck Before Compute 08:04 - Axiom's Moat: Owning Every Layer of the Stack and Generating Proprietary Data 09:27 - Verified Code as Hardware-Software Co-Design and Why Proof Cost Matters 10:45 - Dream, Then Ground: What Is the Riemann Hypothesis of Software? 12:43 - How Intent-Based Programming Keeps a Swarm of Coding Agents Aligned 14:10 - Handling Messy Human Intent With an Adversarial Specification Model 16:58 - Axiom Prover Takes #1 on Lean Eval With a 14 Million Line Submission 18:38 - Where Verified AI Goes in 3 to 5 Years: Deriving Specifications From Code and Data

Madrona Axiom Prover Axiom Math

Chapters

  1. 0:00 Verifying the Value of AI: Mathematical Discovery Meets Verification
  2. 0:53 AI Broke a Math Record Researchers Have Chased Since 2013
  3. 3:28 Why the Stack That Proves Math Theorems Also Verifies Code
  4. 5:07 Why a Million Autonomous Agents Need Formal Guarantees to Build on Each Other
  5. 6:12 Why Agent Swarms Hit a Software Infrastructure Bottleneck Before Compute
  6. 8:04 Axiom's Moat: Owning Every Layer of the Stack and Generating Proprietary Data
  7. 9:27 Verified Code as Hardware-Software Co-Design and Why Proof Cost Matters
  8. 10:45 Dream, Then Ground: What Is the Riemann Hypothesis of Software?
  9. 12:43 How Intent-Based Programming Keeps a Swarm of Coding Agents Aligned
  10. 14:10 Handling Messy Human Intent With an Adversarial Specification Model
  11. 16:58 Axiom Prover Takes #1 on Lean Eval With a 14 Million Line Submission
  12. 18:38 Where Verified AI Goes in 3 to 5 Years: Deriving Specifications From Code and...