An AI strategy can be wrong and still be survivable. The dangerous bets are the ones your product cannot unwind. @dpup, CTO at @thegp, asks founders to separate two kinds of assumptions before making a major AI bet. Some are easy to revise. If token costs fall, you can change how the product works without rebuilding it. Others are structural. If your product is deeply tied to one model provider, leaving may mean rewiring the product. That changes the roadmap conversation. Add a second…