AI
Agentic coding is becoming a staffing decision, not a feature
Three releases this week point the same way. GitHub measures agents starting more than 40% of pull requests, DeepSeek's Harness team published a router-worker prototype built to consume as little human attention as possible, and Google is retiring Jules to concentrate on Antigravity. The common thread is organisational rather than technical: the constraint has moved from whether a model can write a patch to who authorises it, who reviews it and how much supervision a team can afford. Each of those questions now has a vendor-shaped answer.