Changelog Interviews Building the machine that builds the machine
17 snips
Feb 11, 2026 Paul Dix, Co-founder and CTO of InfluxData known for InfluxDB and time-series expertise. He recounts sending AI coding agents on real-world side quests. Topics include agentic coding experiments, verification and QA becoming the bottleneck, designing agent-friendly code and tooling, and building internal systems so agents can produce and validate production-ready code.
AI Snips
Chapters
Transcript
Episode notes
PMs Shipping Prototypes With Agents
- Product managers at InfluxDB used agent tools to prototype UI features without immediate engineering help.
- One PM built a Replit prototype that later became a shipped UI product.
When Agents Miss Load-Time Faults
- Paul let agents make a core change and later discovered subtle failures under load that required manual hand-auditing.
- That led him to pull the plug and personally audit and refactor impacted code by hand.
Agent-First Verification Tooling
- Build comprehensive verification suites and QA tooling that agents can run and validate autonomously.
- Design CLI-based tools and tests so agents can kick off runs, inspect results, and iterate without constant human intervention.