a16zโs David Haber published an insightful article today (link in first comment) highlighting how long-running AI agents first in customer support and more recently in legal are making outcome-based pricing a reality. I bet the same is going to happen for software engineering in the next year. We at Autoheal are building towards this future.
2026's software factories target engineering productivity budgets. I expect 2027's to put self-improvement at their core, enabling pricing against engineering KPIs:
Reliability: Today, investigate incidents and coordinate fixes. Next qtr, learn from resolutions and improve agents, with pricing tied to actual MTTR & incident reduction.
Security: Today, patch vulnerabilities and verify fixes. Next qtr, learn from failed fixes and reopened findings, with pricing tied to verified closure within SLA.
LLM token cost: Today, produce a tested PR. Next qtr, improve agent instructions, model choice and tool use, with pricing tied to token cost per accepted PR.
