Research
AI coding agents ace isolated tests but complete fewer than one in ten tasks in a simulated workplace
A new evaluation places AI coding agents in a simulated developer job with colleagues, project boards and incomplete handoffs, requiring them to keep developing software and reliably close open tickets in a messy daily workflow. Even Opus 5.5 and Astra complete fewer than one in ten tasks fully. The agents finish clear, standalone assignments quickly but lose track when operating as autonomous developers.
AI coding agentsagent benchmarkssoftware engineering