Audit Local AI Coding Logs to Track Your Real Work
Claire Vo analyzed her local Claude Code and Codex logs to reveal coding dropped below 40%. Here is how to run the meta-analysis.
I’m using Jev more than Opus 5.5 or GPT-6. Here’s why.
With Claire Vo · Sunday, October 4, 2026
Claire Vo breaks down TypeSafe AI's decision model Jev, explaining how its low cost and typed output primitives (choice, score, and boolean likelihood) enable high-speed data classification. She shares real-world production use cases, including pairwise pull-request categorization, local coding session meta-analyses, and multi-model pipelines for product insights. Vo also demonstrates how to pair fast decision models with frontier LLMs and real-time voice APIs to build responsive dashboards and interactive applications.
Key takeaways
- Claude Code and Codex store complete user turns and session history directly on your local hard drive. Read more →
- Analyzing 1,700 pull requests across ChatPRD took two minutes and cost 9 cents using fast decision models instead of heavy vector embeddings. Read more →
- Claire Vo built a real-time multimodal app in one afternoon by combining OpenAI's Realtime Voice API, Jev, and the API Ninjas quote endpoint. Read more →
- Standard frontier models generate unstructured text strings, creating high latency, unpredictable output parsing, and steep output token costs. Read more →
- Claire Vo processed 4,500 YouTube comments with TypeSafe AI's decision model Jev, replacing expensive frontier LLM calls with fast, typed classification primitives. Read more →
5 articles from this episode
Claire Vo: How to Audit 1,700 Pull Requests for Nine Cents
Claire Vo explains how pairwise decision models cluster thousands of GitHub PRs in minutes to reveal true engineering effort allocation.
Build Low-Latency Voice Apps with Fast Decision Models
Claire Vo explains how to pair OpenAI Realtime Voice with Jev to build sub-second, emotion-driven interfaces.
Claire Vo: Why Decision Models Cost 4 Cents per Million Tokens
Claire Vo explains how TypeSafe AI's Jev uses typed output primitives to run fast system-one classification at four cents per million input tokens.
How Claire Vo Built Real-Time Search Across 4,500 Comments
Claire Vo used TypeSafe AI's Jev model to classify 4,500 YouTube comments and build a sub-second search dashboard with batch scoring.
More How I AI episodes
- Opus 5.5 is great (but GPT-6 Sol is better)Sep 27 · 6 articles
- Claude is BACK with Opus 5.5Sep 27 · 4 articles
- Stripe built a company brain: Meet KaiSep 13 · 5 articles
- GPT-6 Astra blew away every one of my benchmarksSep 6 · 5 articles