Vercel Ship 2026: Python goes GA, Daemons beta and a pivot to agent infrastructure

Vercel closed its Ship 2026 conference series with a pointed message: the company that spent a decade shaping frontend deployment now wants to be the default infrastructure for AI agents. According to Vercel's event recap, the conference ran across London, Berlin, New York, Sydney and San Francisco, with CEO Guillermo Rauch opening the SF keynote by describing a future of interfaces generated on the fly for whatever task is at hand, agents that reproduce and repair bugs on their own, and software that keeps improving itself using what it observes in production.

A three-part pitch for agentic infrastructure

Rauch framed the strategy in three parts. First, Vercel positions itself as the place coding agents deploy software, exposing more than 200 CLI commands so an agent can operate the whole platform. Second, it is where teams build and run their own agents in production. Third, the platform is automated by agents itself: data from traffic, traces and observability gives Vercel's agents the context to investigate issues autonomously and submit pull requests with fixes rather than merely flagging anomalies. Production context, Rauch argued, is what turns software into a loop, because what happens live determines what gets built next.

Python reaches general availability, Daemons enters beta

CTO Malte Ubl showed that frontends, APIs, backend services, containers, MCP servers and background workers can all run on Vercel, including with custom runtimes and system dependencies. The headline item is Python, now generally available: FastAPI, Flask and Django run on standard CPython with PyPI dependencies, uv and pyproject.toml support, and Vercel Services lets a frontend and backend deploy together with previews of the full application.

Vercel Daemons, launched in public beta, targets continuously running background workers. Teams deploy code or a container image, set the number of instances, and Vercel keeps them running while replacing failed ones — with Celery, Dramatiq and Temporal workers named as supported patterns.

Agent Stack: memory, connectors, budgets and credits

CPO Tomo Occhino walked through updates to the Agent Stack. Eve, Vercel's open-source framework for building agents, gained persistent memory across sessions, extensions for installing tools and skills, and the ability to create and deploy agents from the Vercel dashboard. Vercel Connect is now generally available with more than 1,000 connectors, and MCP Events entered public beta, letting agents react to events from MCP servers while Connect handles subscriptions, renewals and secure delivery.

The AI Gateway gained budget controls at the level of API keys, projects, individuals and teams, plus options for US-only inference and Zero Data Retention so providers do not keep prompts or responses. A new subscription, Vercel Token Pass, is in early access for Pro teams: $30 per month for $80 of token credit on GLM 5.3 and Kimi K3, with more models promised. Vercel Drives, generally available for Sandbox, gives agents persistent workspaces where code, data and saved memory survive across runs even when runtimes or compute change.

Enterprise governance catches up

COO Jeanne Grosser argued that when everyone in a company can build, deployment needs shared controls: who can ship, who can access an app, which systems agents may use and how much they can spend. Two enterprise additions followed. Bring Your Own Cloud on AWS is generally available, running application functions inside a customer's own AWS account and network boundary. Enterprise-Managed Authorization for Vercel Connect entered public beta, letting agents act on employees' behalf through a configured identity provider instead of requiring each person to approve every connection.

Sessions point to where adoption already is

Vercel says more than 2,700 developers, founders and technical leaders attended Ship SF at the Palace of Fine Arts. Sessions included Ora's fleet of test agents, built on Eve, that sign up, integrate and pay on real sites to check whether they work for agents; Anthropic on tuning agent performance across thousands of production sessions with Claude Managed Agents; Meta Superintelligence Labs on building products including Muse; and Slack and Vercel on v, Vercel's internal agent. Adapta detailed a router built on AI Gateway for Brazil's largest generative AI platform, serving 168 models through one integration, selecting a model in under 200 milliseconds and handling more than three million requests a day with automatic failover. A panel with TypeSafe AI's Diogo Almeida and Cerebras' Andrew Feldman covered latency, reliability and cost in fast inference, and Shopify CEO Tobi Lütke closed with a fireside chat, titled The Third Customer, on commerce where agents research and buy on people's behalf.

Why it matters

Vercel built its business on frontend hosting and Next.js; Ship 2026 is a bid to make the platform the home for AI-era workloads — Python APIs, background workers, agents, model routing — sitting next to the frontend. The strategic bet is distribution: as coding agents write and deploy more of the world's software, the platform those agents can operate most easily gains an enormous channel. The product cadence also consolidates what used to be separate vendors — gateway, connectors, persistent storage, spend controls — into one bill. Two caveats apply: this is Vercel's own account of its event, and several of the announced items are still in beta or early access, so the platform shift is more direction than finished state. The enterprise features, notably, show where Vercel thinks the real blockers are: not model capability, but governance, authorization and cost control.