Decision APIs, edge models, and knowing when to stop
OpenAI’s public beta and LiquidAI’s open models focus on bounded decisions rather than long answers. The useful next step is deciding what can run automatically—and what needs a human. This issue also covers practical ways to inspect Internet outages and preview Apple Wallet passes.
Top Stories
Decisions API is in public beta
OpenAI’s public beta returns typed decisions from text and images, with GPT-6 Luna as the supported model.
Why it matters: Classification and routing workflows can request a typed result without parsing a prose answer.
Multimodal open d1 decision models for the edge
LiquidAI’s open d1 models target edge decisions; the smaller multimodal release remains experimental.
Why it matters: Edge-device builders can evaluate a local decision model for bounded tasks instead of generating an answer token by token.
Bridging technical depth and usability: The story behind Radar’s redesign
Radar’s redesign puts traffic and outages on a map, with a clearer path into detailed Internet data.
Why it matters: Teams explaining an outage to less technical colleagues have a clearer starting point for exploring the affected geography.
Apple Pass Designer
Pass Designer combines Wallet pass templates, device previews, field editing, and validation in one tool.
Why it matters: Teams can review a pass’s appearance and required fields before handing the design off for distribution.
Part 3: Knowing when your agent doesn’t know: the confidence layer
The guide connects calibrated confidence, independent review, and human handoff to control when an agent acts automatically.
Why it matters: Agent builders need measured thresholds and an escalation path before allowing uncertain outputs to trigger actions.