Verification, Cost, and Speed: What's Next for AI Coding at Scale
summary
Panel distills practical claims: verification often costs more than generation so teams use AI to produce test harnesses and deterministic checks (recipe runs validate patch files) and treat CI compute and review capacity as real bottlenecks. Scale tactics include hierarchical indexing of a code estate and projection to CSV; one cited estate is ~4.5 billion lines and context windows are ~1M tokens. Budgeting matters — examples: 10 billion tokens in a month and $2k/month per heavy developer — so teams add harness engineering, telemetry, and staged rollouts.
AI-written summary · about
chapters (8)
- 0:00Introduction and panel setup
- 2:10Why AI coding at scale matters
- 3:38Verification techniques for large changes
- 14:34Handling massive codebases
- 21:57Speeding up developer workflows
- 34:44Managing AI token costs
- 47:15Ensuring code accuracy at scale
- 56:24Closing remarks and Q&A
related talks
28mWhy Software Quality Is the Key to Scaling AI Development
51mAI-Assisted Coding: The Future; Between Challenges And Benefits
32mEscape the Try-Again Loop: Verifying AI-Generated Code
39mBack to Basics: Cutting Through the AI-Generated Code Noise
47mCode Is Cheap. Software Isn’t.
19mIf AI Writes the Code, Who Runs Production?
34mBuilding Guardrails for AI Coding Systems
45mAgentic AI Coding Practices for Speed with Quality