9 Economics of vibe engineering: cost optimization at scale

The chapter frames AI cost as much more than an API invoice. It shows how teams often celebrate big savings on token spend, only to discover that hidden costs like developer wait time, retries, overbroad context, and quality erosion are larger than the visible bill. The central lesson is that optimizing the cheapest-looking part of the system can simply move expense into engineering time, lost trust, and slower shipping.

It then lays out a practical optimization ladder: use the right interaction mode for the task, trim unnecessary context, route simpler work to cheaper models, suppress verbose output, and shift non-urgent work to batch or asynchronous execution. The chapter emphasizes that most savings come from a few high-leverage behavioral changes rather than elaborate infrastructure, and that the best optimization is often the one that also improves quality and speed. A recurring warning is that cost-cutting becomes harmful when it pushes the team below a quality or trust threshold, or when the spend is still too small to justify the effort.

Finally, the chapter broadens the economics beyond cloud APIs to local inference, multi-tool portfolios, CI pipelines, and measurement. Local models make sense for privacy, offline use, low-latency tab completion, and some high-volume tasks, but they are not a universal replacement for frontier models. The chapter closes by arguing that teams must measure ROI in developer outcomes, not token counts: cycle time, throughput, first-attempt success, satisfaction, and the ability to identify workflows that should not use AI at all. In that sense, the deepest optimization is deciding where AI genuinely adds value and where it merely burns tokens.