Transcript · 2026-10-23

Welcome to Overnight AI. I'm Alex. And I'm Jamie. We're your AI-generated hosts, and here is your AI news briefing for October 23rd, 2026. Today, we have a new OpenAI API for narrow, fast decisions, a look at the practical work required to deploy AI, a large enterprise rollout, and two Anthropic moves around open agent skills and life sciences research. First, OpenAI has made its Decisions API generally available with GPT-6 Luna. This is not an endpoint for writing an open-ended answer. Developers define a question, give it allowed answers and context, and get back either a probability, a choice, or a score. That distinction matters. The intended jobs are things like routing a request, classifying input, or choosing whether an agent should continue, retry, or ask for clarification. OpenAI says the API accepts text, images, and native JSON, and can now handle up to 400 questions in one request. OpenAI also says the endpoint is up to 10 times faster than using GPT-6 Luna through its Responses API, with inference residency in the United States and European Union. Those are OpenAI's own performance and deployment claims, not a broad finding about model quality. The listed price is 10 cents per million input tokens, with a separate 1 cent cached input price, and no output token charge. As always, the practical question is whether a narrowly specified model decision is reliable enough for the action attached to it. That brings us to OpenAI's separate account of its forward deployed engineering team, or FDE. The group works alongside customers to make AI systems function in production, then feeds the gaps it finds back to product and research teams. OpenAI describes an unnamed semiconductor engagement where it says AI-assisted directed test generation saved about 48% of time, and kernel optimization improved median performance by 22%. But the more durable point is its warning that a strong benchmark result does not automatically become a reliable real-world system. Right. The company says evaluations, integrations, operational constraints, and feedback from users can be the actual work between a capable model and deployment. Those customer examples are OpenAI's account of its own program, so they are useful evidence about its approach, not independent proof of universal results. And that deployment theme is visible in Uber's company-wide ChatGPT Work rollout. Uber says more than 30,000 employees will gain access after an initial 1,000-person pilot. It calls this its first company-wide agentic platform tailored for knowledge work. According to OpenAI, ChatGPT Work can connect with Google Drive, Gmail, Google Calendar, and Slack for research, meeting briefs, presentations, data analysis, and multi-step workflows. GPT-6.1 Sol will be the default capability, and Uber says it is pairing the platform with an internal skills library, approved integrations, and training. Neither company is claiming a finished company-wide productivity result yet. They say they will share more about adoption, use cases, and measurable outcomes as the rollout continues. That's worth keeping separate from the fact of the rollout itself. Next, Anthropic has donated Agent Skills to the Agentic AI Foundation under the Linux Foundation. Agent Skills is an open specification for a folder containing a SKILL.md file and any scripts or reference material needed to teach an agent a particular task or domain. Anthropic says more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot, and VS Code, have integrated the standard since it was published in December 2025. The company also says 1 million skills have been published and nearly 280 million installed. The donation includes the Agent Skills name and logo and its GitHub organization. Engineers from Anthropic, Google, OpenAI, and other companies will maintain it as individuals. Anthropic has proposed versioned releases, a formal proposal process, and security guidance, but those are a roadmap, not yet settled governance rules. Finally, Anthropic opened a call for life sciences researchers to propose special projects. Principal investigators and groups at academic institutions, nonprofits, and government laboratories can apply to work directly with Anthropic researchers. The company says selected guest investigators would spend 8 to 12 weeks on work in causal human biology, structural and computational molecular biology, medicinal chemistry, drug discovery, or clinical science. Anthropic says it will run models and harnesses at no cost, while collaborators retain rights to their data and independent follow-on research. Anthropic says the collaboration could publish outputs and open source accompanying data and code, but it is looking for preliminary evidence that can be produced within that relatively short period. That is not a claim that an AI-assisted project can settle difficult biological questions by itself. Today's throughline is narrower than a model launch. Systems are being packaged into bounded decisions, enterprise workflows, and shared standards, while the harder questions of validation and accountability remain attached to the work. For the original announcements and the qualifications behind them, find the source links in the show notes. Thanks for listening to Overnight AI.