YC CEO says he ships 37K LoC/day AI code. A developer looked under the hood
THE SO WHAT
Claims of 37K lines of AI-generated code per day—and the subsequent scrutiny—highlight that raw output metrics are now more PR than productivity. For engineering leaders, the only numbers that matter are defect rates, review time, and cycle time on shipped features, not how much code the model can spit out.
READ THE SOURCE
MORE FROM THE WIRE
Applied AINearly all retailers have now implemented AI, but many are still waiting to see business value
Retail is now a live case study in AI theater—near-100% adoption, but only half seeing ROI because legacy systems and dirty data choke the pipeline. If you own a P&L, stop greenlighting new AI pilots until you’ve funded data quality and integration work on the core stack.
Applied AIAI only creates value when it fixes the workflow
The bottleneck isn’t model quality, it’s process design—AI dropped into broken workflows just accelerates the mess. Before your next AI build, map the end-to-end workflow and decide explicitly what changes, who reviews what, and where decisions are made.
Midjourney founder says new AI coding tools are leaving his friends more productive — and 'extremely drained'
AI that boosts output while burning out developers is a productivity mirage—throughput goes up, sustainable capacity goes down. Track not just velocity metrics but error rates, rework, and burnout signals if you’re pushing AI coding tools into your teams.
Applied AIUS cyber agency is using Anthropic’s Mythos to audit government code, sources say
If CISA is quietly pointing an offensive-grade model at federal code, AI-assisted red teaming is moving into the core of state cyber defense. Enterprises should assume the same tools will be used—by both attackers and auditors—against their own software and adjust threat models accordingly.