Unify CTO shares how they cut 95% of AI agent costs in two weeks
hwchase17 · x · 2026-08-13
The Max Agency podcast hosted Connor Heggie, CTO and co-founder of Unify, to share how his team managed to cut 90-95% of AI agent costs just two weeks before their product launch.
Key Engineering Insights:
- Prompt Caching Limits: Discusses the hidden 15-requests-per-second ceiling inside OpenAI's prompt cache and strategies to optimize cache hit rates.
- Subagent Architecture: Explains why their subagents are designed as simple function calls, comparing fork versus child subagent models.
- Evaluation & Judging: Emphasizes that the LLM judge must be from a different model family, sharing how his background in self-driving influences their current agent eval workflows.
More from coding & agent
- ProtonGraph: Open-Source Node-Based Tool for Procedural 3D Content Generation — tom_doerr · 2026-08-14
- Forensics Tool bulk_extractor Fixes 100+ Issues with Codex Aid — matthew_d_green · 2026-08-14
- New Paper on Coding Agents: Beyond Clones Towards a Long-Term Perspective — teortaxesTex · 2026-08-14
- AI Agent Setup: Isolating Global and Project-Level Instructions Across Three Directories — brandon_galang · 2026-08-14
- Claude Code cuts dependency upgrades from 6+ hours to 85 minutes with zero regressions — JeremyCMorgan · 2026-08-14
- Introducing ReactBench: A Frontend Coding Benchmark Beyond Passing Tests — aidenybai · 2026-08-14