November 2026 Summaries

Vercel AI Gateway’s November 2026 Production Index, based on anonymized aggregate traffic through October, reports rapid changes in production AI adoption across more than 100 trillion monthly tokens. Decision models, which classify or select rather than generate text, reached 26% of requests within weeks of their introduction but represented only 0.53% of spending because of their low cost, with adoption especially high among the largest users. Voice-model adoption rose from 4% of paid teams in July to 11% in October as teams increasingly added speech-to-text and text-to-speech capabilities to existing AI applications. Anthropic and OpenAI reached nearly equal shares of paid teams, although Anthropic attracted almost twice as many new teams in October, while OpenAI increased its portion of gateway spending from roughly 9% to nearly 15% since December. Model selection has become highly fluid, with over half of language-model requests going to versions under 60 days old and many teams quickly shifting to newer releases or competing labs. The report also notes record adoption of image and video models, strong usage of Anthropic’s models, and leading spend positions for Opus 5.5 in language, OpenAI’s GPT-Image-2.5 Sunburst in image generation, and ByteDance’s Seedance 2.5 in video.

Vercel Schedules, currently in public beta, extends traditional static cron jobs by enabling applications and AI agents to create, modify, and cancel one-time or recurring tasks at runtime without redeployment. Schedules specify when work should run and where it should be delivered, while application code performs the task, supporting static configuration through vercel.json or the Build Output API as well as dynamic management through a JavaScript SDK, CLI, or API. Tasks can invoke private functions, HTTP paths, or Queue topics, with preview and custom-environment testing available before production. An example shows an AI agent scheduling a future request to summarize Hacker News stories, with Vercel Queues delivering the stored request to the application when due so it can validate permissions and execute the agent. The service also supports recurring tasks such as daily cleanup jobs, including private function invocation and built-in verification that requests originate from Vercel’s scheduling service.

Vercel has launched Security Review in public beta as a Vercel Agent capability for Pro and Enterprise teams that scans code pushed to connected GitHub repositories for vulnerabilities, including authorization flaws and exposed secrets. Triggered on every push and scoped to changed code relative to a deployment or branch base, it reports findings in Vercel’s Security dashboard, through the CLI, and, for private or internal repositories, as GitHub checks and pull-request comments. The agent examines code context and reachability, applies customizable plain-language security policies, deduplicates and tracks findings across deployments, and can validate, implement, and test minimal fixes when authorized. Teams may also retrieve findings through the CLI or copy them in Markdown or JSON for use with other coding agents, while Enterprise users can export audit events to SIEM systems. Vercel plans to extend support to CLI and v0 deployments, non-Vercel repositories, and pull-request fix suggestions; reviews are free until November 10, 2026, after which usage follows Vercel Agent and token-based pricing.