An interesting implication of your second category: evals are becoming the moat. If everyone ships on the same frontier model, the teams that figure out how to encode their taste and requirements into automated scoring will compound quality gains over time. Teams that skip this

→ View original post on X — @randal_olson