Vercel Ship 2026: Agent Infrastructure Goes Fully Production-Ready

Vercel spent a decade shaping how the web is built. At the just-concluded Vercel Ship 2026, the company announced it intends to bring the same methodology to the world of AI Agents. The tour covered London, Berlin, New York, Sydney, and San Francisco, with thousands of developers attending the events.

In his opening keynote in San Francisco, CEO Guillermo Rauch offered a clear verdict: the companies that win the next decade will be built from the ground up on infrastructure "designed for Agents." The key ingredient in that infrastructure is "production context" — everything that happens in production drives what gets built next.

Agentic Infrastructure: Vercel's Three-Layer Strategy

Rauch broke Agent infrastructure down into three layers: Vercel is where coding Agents deploy software, Vercel is where you build and deploy your own Agents, and Vercel itself is operated by Agents.

The first layer is most directly embodied by over 200 CLI commands. When you ask a coding tool "where should I deploy," the answer points to Vercel, because the entire platform interface is designed around how Agents work. The second layer provides all the tools needed to build and scale production-grade Agents. The third layer is especially noteworthy: Vercel runs your applications, handles traffic, tracing, observability, and anomaly detection — and that data feeds back to give Agents the context they need for autonomous investigation, ultimately producing not just alerts but Pull Requests that can be merged directly.

Full-Stack and Agent Deployment Capabilities Upgrade

In his keynote, CTO Malte Ubl demonstrated how frontends, APIs, backend services, containers, MCP servers, and background tasks all run unified on Vercel. Several key announcements stand out:

Python is now generally available (GA). Standard CPython, PyPI dependencies, uv, and pyproject.toml are supported, and FastAPI, Flask, and Django all run. Paired with Vercel Services, you can deploy frontend and backend together and still get full preview environments.

Vercel Daemons enter public beta. This is a continuously running background Worker solution — deploy code or a container image, set the instance count, and Vercel keeps it running and replaces it on failure. Celery, Dramatiq, Temporal Workers, or existing processes can all be plugged in directly.

Agent Stack: A Complete Pipeline from Build to Production

CPO Tomo Occhino presented the latest progress on the Agent Stack. The open-source Agent framework eve adds cross-session persistent memory, extensible installation of tools and skills, and the ability to create and deploy Agents directly from the Vercel Dashboard.

Vercel Connect is now generally available, with over 1,000 connectors; MCP Events enters public beta, letting Agents respond to MCP server events, with subscription, renewal, and secure delivery handled by Connect. AI Gateway adds budget controls at the API Key, project, personal, and team levels, along with US-only inference and zero data retention options. Vercel Token Pass opens early access to Pro teams: $30 per month gets you $80 worth of GLM 5.3 and Kimi K3 token credits. Vercel Drives (Sandbox) is generally available, providing persistent workspaces for Agents — code, data, and memory survive across run cycles, and switching runtimes or compute resources requires no environment rebuild.

MCP (Model Context Protocol) is an open protocol introduced by Anthropic in late 2024 to let AI models connect to external tools, data sources, and services in a standardized way. Think of it as "USB-C for the AI world": previously every Agent framework had to write separate adapter code to connect to GitHub, Slack, databases, and other systems; MCP abstracts these capabilities into a unified Server/Client interface. The connectors and MCP Events provided by Vercel Connect are built on this protocol — the former lets Agents access third-party services through 1,000+ ready-made connectors, while the latter lets Agents respond to events pushed by external MCP servers. Subscription, renewal, and secure delivery are hosted by Vercel, so developers don't have to handle OAuth refresh and Webhook retry logic themselves.

AI Gateway is Vercel's unified entry point for model calls, converging the APIs of multiple model providers — OpenAI, Anthropic, Google, GLM, Kimi, and others — into a single endpoint. Developers can switch models through a single integration, with the gateway handling routing, retries, caching, and usage statistics. The newly added budget controls support four granularities — API Key, project, personal, and team — meaning teams can set a separate monthly cap for an experimental Agent to avoid runaway calls producing unexpected bills. The US-only inference and zero data retention options target enterprise customers with hard data compliance requirements.

Enterprise-Grade Controls and Real-World Case Studies

COO Jeanne Grosser focused on "what happens when everyone in a company can build." Engineers, marketers, analysts, and salespeople need a common deployment platform, along with controls over release permissions, application access, systems available to Agents, and spending limits.

Vercel for Enterprise adds two capabilities: Bring Your Own Cloud on AWS is generally available, with application functions running inside your own AWS account, within your security and network boundaries; Enterprise-Managed Authorization for Vercel Connect enters public beta — once supported services and identity providers are configured, Agents can act on behalf of employees without authorizing each connection individually.

BYOC (Bring Your Own Cloud) is a deployment model between pure SaaS and fully self-hosted: the control plane (Dashboard, CI/CD, observability, etc.) remains hosted by Vercel, but the functions and containers that actually run business logic run in the customer's own AWS account. This way code, secrets, and network traffic never leave the enterprise's existing security and compliance boundaries, and you can directly reuse existing VPC, IAM policies, audit logs, and firewall rules without giving up Vercel's developer experience. For regulated industries like finance, healthcare, and government, this is often a prerequisite for adopting a platform at all.

Enterprise-Managed Authorization solves the identity problem when Agents act on behalf of employees within an enterprise. Traditional OAuth authorization requires each user to consent individually to each connection's permissions; when a company deploys hundreds of Agents involving dozens of SaaS services, this per-person, per-connection authorization process quickly becomes a burden. Enterprise-managed authorization lets administrators centrally configure at the organization level which services and which identity providers (such as Okta, Entra ID) can be used by Agents, after which Agents can call these services on behalf of employees without going through separate authorization. This elevates permission governance from "each user decides for themselves" to "IT department unified policy" — a necessary condition for deploying Agents at enterprise scale.

Practical Insights: From Anthropic to Shopify

Ship San Francisco drew over 2,700 developers. Several talks provided noteworthy real-world data:

Prashant Ratanchandani of Meta Superintelligence Labs and Tomo discussed the experience of building products like Muse on Vercel. Vercel's Matt Lewis and Slack's Mala Neti explained how the internal Agent "v" was built around Vercel's own data. Max Peters of Adapta broke down the routing behind Brazil's largest generative AI platform — built on AI Gateway, it serves 168 models through a single integration, completes model selection within 200 milliseconds, and handles over 3 million requests per day.

The London stop was equally packed. Vercel's Matan Kushner revealed that the support Agent Vertex now automates 91% of support tickets, saving 5,000 engineer hours per month. Abhi Sivasailam introduced the data Agent "d0," which already sees 45% of its questions coming from other Agents rather than humans. Tomas Jansson of Currys showed the three-phase evolution of a Nordic retailer's e-commerce stack, with the Next.js migration cutting time to first byte by 40%.

Closing Fireside Chat: The Next Step in Agent Commerce

Rauch's conversation with Shopify CEO Tobi Lütke centered on "the third customer," exploring a world where Agents research, compare, and purchase products on behalf of humans. Lütke proposed the concept of "project-based commerce" — planning Halloween decorations, doing a 3D printing project with your kids, then ordering all materials at once to arrive before the weekend. These are things people never had time to do, and Agents can do them for them.

For traditional storefronts, Lütke believes there is still room for growth in unique, carefully designed shopping experiences. What will truly struggle are merchants offering undifferentiated experiences. On the engineering side, Shopify has redefined its role as a "verification engineer" — building test frameworks and acceptance tests to make changes verifiable. They are rewriting their storefront renderer from Ruby to Rust, comparing output byte-for-byte before switching. The internal Slack Agent "River" dominated the code deletion leaderboard during a hackathon.

An opportunity worth developers' attention: Shopify's Catalog API opens up merchant product access, letting you build new commerce experiences without first recruiting both sides of a marketplace. Product discovery, recommendation, and curation are real business opportunities.

Agentic Commerce refers to the entire process of AI Agents completing product research, comparison, selection, and purchase on behalf of humans. Its key difference from traditional e-commerce is that the initiator of the purchase decision shifts from a human browser to an Agent, and product display, search ranking, and checkout flows all need to be redesigned for machine consumers. This means merchants need to provide structured, machine-readable product data (price, inventory, specs, reviews), not just marketing pages optimized for human eyes. The "project-based commerce" Lütke described is a typical scenario of this trend — a user describes a goal (like "Halloween decorations"), and an Agent decomposes the requirements, compares prices across stores, and places a single order. For developers, whoever can provide a stable, low-latency, semantically clear catalog API could become the infrastructure layer of the Agent commerce era.

One Signal

From 91% support ticket automation to the automated merging of 60-70% of low-risk PRs, the signal from Vercel Ship 2026 is clear: Agents are no longer demos, but engineering infrastructure entering production environments. Identity, permissions, and verifiability are the key variables determining whether Agents can truly go live.

One-line Summary

Vercel Ship 2026 unveiled a three-layer Agent infrastructure strategy and full-stack upgrades, marking Agents' entry into production environments.

Paragraph Summary

Vercel Ship 2026 toured London, Berlin, New York, Sydney, and San Francisco, with the core message that Vercel wants to bring its web-building methodology to AI Agents. CEO Guillermo Rauch proposed a three-layer Agentic Infrastructure: a platform where coding Agents deploy software, a platform where developers build and deploy their own Agents, and Vercel itself operated by Agents. Technical releases include Python GA, Vercel Daemons public beta, the eve framework's persistent memory, Vercel Connect GA, AI Gateway budget controls, and Token Pass early access. On the enterprise side, BYOC on AWS and Enterprise-Managed Authorization were introduced. In real-world data, Vertex automates 91% of support tickets, and the d0 data Agent sees 45% of questions coming from other Agents. In the closing conversation, Shopify CEO Tobi Lütke explored the future of Agents shopping on behalf of humans.