Why AI Has Not Made Enterprise Software Cheaper

Summary

Gal Ratner argues that AI has made code generation far faster without making enterprise software cheaper. He points to Anysphere, whose rapid Cursor revenue growth was accompanied by inference costs reportedly consuming 40% to 70% of each revenue dollar, prompting repeated shifts toward usage-based pricing. Across the market, Zylo reported that 79% of IT leaders faced a renewal price increase, 78% encountered unplanned consumption or AI charges, and 61% cut or paused a project because of unexpected software costs. The article says productivity gains have historically been absorbed by expanded scope: buyers now expect security, integrations, mobile clients, real-time features, compliance, and reliability that were not part of earlier software products. At the model layer, capability prices have fallen sharply, but cheaper intelligence has encouraged more inference, consistent with the Jevons paradox; the article cites rising OpenAI inference costs and enterprise API spending as evidence. Agentic systems add a variable cost of goods sold because they repeatedly read context, plan, execute tools, run tests, and retry, making per-seat pricing harder to sustain. Ratner concludes that commodity build work may approach zero in price, while integration, migration, compliance, data quality, and architecture remain valuable. In his view, enterprise software is becoming a metered services business whose greater productivity can support broader work and higher bills rather than automatic price reductions.