Vercel AI Updates: July 22, 2026

1. Vercel Launches The Vercel Agent For Production Operations

Vercel. Vercel introduced the Vercel Agent, an AI assistant that monitors production environments and performs tasks such as investigating incidents, reviewing pull requests, tracing cost increases, and fixing broken builds, all with human approval. It runs under a separate vercel-agent identity so its actions are traceable, requests permissions scoped to a specific proposed plan rather than inheriting broad access, and executes generated code inside isolated Firecracker microVMs. Combined with Vercel’s immutable deployments, the design lets the agent act autonomously while keeping mistakes reversible, for example recommending an instant rollback when it traces a failure to a recent deploy. Source

2. AI Gateway Adds Service Tiers For Cost And Latency Control

Vercel. Vercel added service tiers to the AI Gateway, letting developers trade off cost and latency per request across default, priority, and flex tiers. Priority processing runs faster at roughly 1.8 to 2x default pricing for interactive workloads, while flex targets background jobs at about half the cost with potentially higher latency. Tiers are set via serviceTier under providerOptions.gateway, work across the AI SDK, Chat Completions, and Anthropic Messages formats for OpenAI and Gemini models, and fall back to the default tier and rate when a tier cannot be applied. Source

3. Gemini 3.6 Flash And 3.5 Flash-Lite Arrive On The AI Gateway

Vercel. Vercel made Google’s Gemini 3.6 Flash and Gemini 3.5 Flash-Lite available on the AI Gateway. Gemini 3.6 Flash targets coding, agentic work, and web development with lower token consumption and fewer model calls, while Gemini 3.5 Flash-Lite adds agentic capabilities aimed at subagents that handle scoped parts of a larger task. Developers can reach them with the model identifiers google/gemini-3.6-flash and google/gemini-3.5-flash-lite. Source

4. Poolside’s Laguna S 2.1 Coding Model Lands On The AI Gateway

Vercel. Vercel added Poolside’s Laguna S 2.1, an open-weight Mixture-of-Experts model with a context window up to 1M tokens that runs in thinking and no-thinking modes, to the AI Gateway. The model targets agentic coding and long-running projects, reporting 70.2% on Terminal-Bench 2.1, 78.5% on SWE-bench Multilingual, and 59.4% on SWE-Bench Pro in thinking mode. It launches in two versions, a free tier with 256K context and a paid version with the full 1M-token context. Source

5. Vercel Connect Adds More Than 90 Preset Connectors

Vercel. Vercel Connect added preset connectors for more than 90 external services, including Shopify, Okta, Workday, Jira, and Sanity. The presets are predefined configurations that pre-populate branding and authentication details so teams skip most manual integration setup, then request scoped tokens through the getToken function using a connector identifier. Unlike managed connectors, presets do not automatically register an application with the external service, so developers handle that step themselves. Source

6. Vercel MCP Can Now Complete Purchases

Vercel. Vercel extended its MCP server so AI assistants can complete transactions rather than only retrieve information, quoting a price, explaining whether a charge is one-time or recurring, and finishing a purchase only after the user confirms. Supported actions include team plan upgrades, prepaid credits for v0 and the AI Gateway, the enterprise SIEM add-on, and domain registration. Transactions require appropriate team permissions and a payment method on file. Source