Claude Code picks Resend over SendGrid roughly 9:1. Resend launched in 2023. Twilio paid $3B for SendGrid in 2019. Nobody at Resend made this happen. Nobody at SendGrid stopped it.

Agents don't evaluate products. They pattern-match to training data. Every GitHub repo, Stack Overflow answer, and tutorial published before the training cutoff is an acquisition channel — running 24/7 across every model deployment worldwide. Supabase grew 1M → 4.5M developers in 12 months not because they out-marketed anyone.

Vercel's CEO explained what happened to Neon — technically excellent, arguably better serverless architecture — when Databricks acquired it: "Agents favor infrastructure well represented in training. Not reinventing the wheel with new query languages was rewarded."

I ran growth at W&B for 7 years and wrote up what this means for PLG growth — what's changed at each stage, why token-to-value is the new time-to-value, and the three compounding loops that make this a flywheel rather than a funnel.

Part 1 (the agents flywheel): https://improbabilityvc.substack.com/p/the-agent-flywheel

Part 2 (implmentation checklist): https://improbabilityvc.substack.com/p/the-agent-flywheel-ch...