Chris Potts introduces AI coding CPI

While LLM unit costs are plummeting, tokenflation means developers are actually getting less raw code per token because modern reasoning models use more conversational context and internal thought tokens.

• –**The "Tokenflation" Reality:** Raw volume metrics like PRs, files touched, and lines of code produced per token have dropped by up to 85% in just over two months.
• –**Hedonic Adjustments Matter:** The 5% improvement in code survival rate (from 90% to 95%) is the silver lining, indicating that models are producing more robust, long-lasting code rather than sheer volume.
• –**Agentic Workflows are Expensive:** This index confirms that agentic execution and deep-thinking frameworks will shift developers from buying raw token volume to buying verified outcomes.