Spotify just published something every founder burning money on coding agents should read. One engineer cut Claude Code token use by 90 percent. Not by writing less code. By stopping the frontier model from doing grunt work. Most of what an agent does is not thinking. It is reading five files to answer one question, or writing the twentieth test that looks like the other nineteen. That burns expensive tokens on I/O. Spotify routed bulk reads and boilerplate to Gemini 2.5 Flash and kept Claude for the hard problems. Gartner says AI coding costs will pass the average developer salary by 2028. A quarter of engineering leaders already spend USD 200 to USD 500 per developer per month on tokens. Some are past USD 2,000. The seat license is not the bill. The bill is feeding a frontier model work a cheap model can finish. Route the workload or the token invoice owns you.