Why this is in the vault
Direct on-the-ground consumer-agent evidence (Stern's back-to-school-shopping agent test, hair-appointment booking failure) that names the exact bottleneck — web/login compatibility, not model capability — blocking agent adoption for normal users.
The core argument
Ben Thompson interviews Joanna Stern (WSJ alum, now independent via New Things and YouTube) on two threads: the iPhone Duo foldable and consumer AI-agent readiness. On the Duo, Stern's verdict is cautiously positive on form factor (Apple matched Samsung's Z Fold 8 sizing, added polish like instant external-to-internal screen continuity and a repositioned dock) but flags three real risks — price, battery life, and whether the internal screen ends up an unused "parlor trick" if most usage stays on the external display. She reports Apple showed almost no work for how unsupported third-party apps degrade on the new form factor beyond "a gap on the right-hand side."
On agents, Stern describes concrete testing: Meta's Muse agent failed to book a hair appointment because it couldn't push through account login as a guest; a multi-agent back-to-school shopping test (from her own newsletter) showed real year-over-year improvement in following PDF instructions, but every agent still stumbled on retailer web compatibility (one agent picked Target, gave up on quality mid-task, and pivoted to Walmart on its own). Her framing: the blocker for normal-person agent adoption isn't model reasoning, it's the open web's assumption of a logged-in human clicking buttons. Both Thompson and Stern close on a shared belief that "customers don't change their habits" — successful AI products cater to existing behavior rather than requiring new ones — and that their own optimism about agents stems from a generational love of computers as tools, not from believing hype-cycle claims that computers should disappear into an invisible agent layer.
Mapping against Ray Data Co
Directly reinforces the /investing:label-historical-phases and agent-adoption tracking thread (see 2026-06-02-every-eight-levels-ai-adoption): Stern's shopping-agent test is a rare falsifiable, repeated-year data point on where consumer agent capability actually sits — "better at following instructions, still blocked by web plumbing" — which is the same instrumentation gap RDCO's own agent-harness work (Workflow scripts, browser automation via Playwright) has to solve for any client-facing agent, not just consumer ones. The "customers don't change their habits, cater to them" framing is also a useful check against any RDCO pitch that assumes clients will restructure their process around an agent rather than the agent absorbing the existing one.
Related
[[2026-09-10-stratechery-iphone-duo-intelligent-personal-hub]] [[2026-06-02-every-eight-levels-ai-adoption]] [[2026-07-20-every-ai-workflow-adoption]]