[Editorial] Chinese X is arguing one thing: after chat, agent capability is outrunning trust

Why it matters

Our take: Chinese X threads that look separate—Zhipu’s Tang Jie saying the chat war is largely over and the next fight is memory/long-horizon agents, Jev’s “decide don’t chat” split, and ZCode’s alleged silent workspace+git upload that triggered corporate bans—are one crack: after chat’s diminishing returns, the battlefield is executable agents, but capability narratives are outrunning trust. Tang’s path is the vendor consensus; Jev shows teams want cheap decision steps off frontier generators; ZCode forensics (ferstar et al.) and company bans nail the permission boundary. The next divide is not only who codes better, but who makes local access, upload, and key policy auditable, disableable, and trustworthy by default. In-house editorial synthesis of public discussion, not a vendor brief.