A Better Model Ships. Do You Switch?

You are the CTO and a co-founder of an AI-powered customer support startup. ~$1.2M ARR, 18 months in, Series A close is six weeks out. Your entire production stack runs on GPT-4 — your prompts are GPT-4-tuned, your evals are GPT-4-benchmarked, your enterprise contracts cite "GPT-4" by name in the SLA.

This morning, Anthropic announced Claude 5: meaningfully better at instruction-following on internal benchmarks, 40% cheaper per token, 300ms faster median response time. Twitter is on fire. Two of your enterprise customers have already DM'd asking if you'll move. Your Series A lead investor sent a Slack: "Are you switching?"

Engineering capacity is the constraint: you have 3 engineers, all shipping the product roadmap that the fundraise is based on.

The principle: Don't pivot a production stack on a model release. The right time to switch foundation models is when your wrappers and benchmarks tell you to, not when the press release does. Hype-driven migrations cost more in eng time than the marginal capability buys back in user value.