Friday Five

This week's through-line: the industry keeps shipping systems that assume identity and comprehension rather than verify them. A GP surgery's AI receptionist could not parse a stroke survivor's speech; Kremlin operatives put fabricated surrender orders in the mouths of Ukrainian officials; a synthetic Bill Gates pitched crypto to solve the water crisis. The same week, vendors sold agentic media buying and cheaper long-context inference — capability scaling on one side of the ledger, with no matching investment in confirming who is speaking or who is being understood.

Scout's Pick — Outlier

Consumer

AI receptionists seem like a dangerous idea after a patient revealed her doctor’s system could not understand

The failure surfaced only because a patient with aphasia spoke up publicly. The surgery had no mechanism to detect comprehension breakdown, and disclosure came from the person harmed rather than the system.

Culture

AI Puts Words in the Mouths of Ukrainian Leaders

Fabricated surrender orders attributed to named officials exploit the same gap: distribution infrastructure treats a voice as its speaker, with no verification layer between utterance and audience.

Brand

Agentic AI will Put Media Budgets In Motion: Dentsu X’s Brad Stockton

Media budgets moving on agent authority extend spending power to systems whose instructions arrive unauthenticated. Capability grows while the question of who issued the order remains unexamined.

Connected

AI inference gets a new tier as context windows grow

Cheaper long-context inference lowers the cost of generating plausible speech at length, enabling volume scale without corresponding investment in verification at the point of consumption.

AI

Meet Digital Experience Tech’s New Interface: Agentic AI Systems Of Action

Systems of action convert interpretation into execution, eliminating the moment when a human might catch a misheard request or misidentified requester.

6 themes · 160 signals · 80 sources