Steve Jamison reports that Jellyfish's observed AI spending grew sixtyfold while merged pull requests per engineer rose 43% and combined salary-plus-AI cost per merged pull request fell 26%. The proposed cost model treats tokens as one input to engineering delivery rather than an isolated invoice; it spreads salary across completed work instead of demonstrating equivalent cash savings.
Steve Jamison compares organizations grouped by AI spending, then compares individual developers' heavier agent-skill weeks with their own lower-use weeks. Those associations suggest that more output can outweigh higher token costs, but the talk does not establish that spending alone caused the differences or that every organization should spend more.
Steve Jamison uses reverts as one limited quality signal and notes that pull requests have not simply become smaller. More lines of code and merged changes do not by themselves measure customer value, maintenance burden or successful outcomes. He advocates tracking total cost, throughput and quality together rather than choosing an AI budget from token usage alone.
Watch on YouTube
