AI investing needs more demand side obsession. The supply-side stuff is pretty well mapped right now and its just so point in time shortage focused while the underlying tech is a deflation machine. Think its kind of important this is cost driven tech the customer expects to deflate annually from workload standpoint because of computing efficiecny gains and all kinds of algo/workload optimizations. This isn't like saas where the cost is hidden and all the compute efficiency gains go to the vendor. No sfdc customer was getting ec2 pass through price cuts. Where we at today an agent is still being viewed more as human in the loop tool then full labor and as a tool the expectations will be for the cost to deflate notably as the workload matures . If it doesn't then its gonna be compared to labor and figuring out that threshold in coding is probably worth some effort as it probably benchmarks all that follows. I'd say sub 10% is tool and over 20% enters wages convo land and in between is no man's land. Figure u are chasing a bunch of curves. Frontier capability curve (inflating), compute curve (still deflating), optimization curve(deflating), adoption curve etc. If the capability gets step function better obviously ur looking at the cost vs salary. And if ur still at the same capability and content w ur engineers agentic enhanced output, you will expect the tokens for the same work to drop notably. U.S. soft engineers probably $400-$500bl wages so figure $100bl in token spend is the threshold for where scrutiny starts. Digging into that question at customer level is what i'd like more answers too.