The most advanced branch of the timeline: autonomy pays off fastest here because the result is immediately checkable by tests.
About this eventThe industry measure: the share of real issues from open repositories closed without a human.
SWE-bench tests a real issue in an existing repository with tests and dependencies rather than generation of an isolated function. The Verified set is additionally filtered by people, but a score is still not equivalent to engineering work: tests may miss architecture, security and maintainability.
Source: SWE-benchAbout this eventSystems plan and test complete changes instead of completing lines.
About this eventTriaging failed builds, fixing tests and bumping dependencies moved into the automated loop.
About this eventWriting code became cheaper than reviewing it — a bigger shift than generation itself.
About this eventCode nobody read accumulates faster than before. The industry has not yet seen what that costs in three years.
About this eventMoving from closing tickets to owning architecture: choosing trade-offs rather than executing decisions already made.
About this eventA product defined by requirements and tests, with code regenerated automatically and read by nobody.
Primary material behind this dossier: papers, lab publications and official reports.