SHIP SF Opening Keynote

Highlights

• Deployment velocity reached 1 billion deployments every four months on Vercel.

• Deployments triggered by coding agents increased 28 times over the past 12 months.

• Standard CPython is generally available on Vercel, supporting runtimes up to 30 minutes per request and continuous execution across hours or days.

• Vercel Daemons introduce an always-on serverless compute primitive to replace custom Kubernetes management for background workers like Celery and Temporal.

• AI Gateway serves over 4 trillion tokens per day across hundreds of models with zero data retention and optional US-only routing.

• Vercel Drives provide persistent storage that can be mounted across Firecracker microVM sandboxes, enabling stateful agent handoffs and shared dependency caching.

• Bring Your Own Cloud on AWS is generally available, allowing enterprises to run the full Vercel infrastructure inside their own cloud accounts.

Timeline

1. 00:00 - 08:06

   Agentic Infrastructure and the Production Feedback Loop

   Software development is shifting toward autonomous loops where systems build, test, and repair themselves using real-time production signals. Just as modern race car engineering uses real-time telemetry from every lap to refine performance before the finish line, development platforms feed runtime errors directly back into coding agents. The platform supplies agents with full orchestration primitives, turning them into automated factories that handle deployments, infrastructure management, and optimization pull requests autonomously.

   • End-to-end software factories can deploy on Vercel in under three minutes.
   • Coding agent deployments surged by 28x while developer count tripled over the last year.
   • Autonomous code remediation requires production context such as runtime 500 errors and UX telemetry.
   • Over 200 platform CLI commands allow coding agents to provision sandboxes, configure routing, and purchase domains directly.
2. 08:07 - 15:20

   Full-Stack Backend Expansion: Python, Containers, and Daemons

   Backend services require compute primitives beyond stateless front-end requests, which process 2 trillion hits monthly. Native Python support removes the need for architecture rewrites, allowing frameworks like FastAPI and Django to execute long-running jobs alongside TypeScript queues. For custom dependencies and legacy stacks, direct Dockerfile execution brings managed previews and rollbacks to arbitrary languages. The launch of Vercel Daemons fully eliminates raw Kubernetes cluster management by providing always-on, auto-recovering background instances for persistent subscribers and queuing workloads.

   • Native CPython runtime support is generally available with standard PyPI dependencies and single-commit unified previews.
   • Container deployments run Dockerfiles with automatic scaling, rollbacks, and previews across languages like Go, Rust, Ruby, and Java.
   • Vercel migrated over 25 internal API microservices from Kubernetes to Fluid Compute, serving 1 billion internal API requests monthly.
   • Vercel Daemons entered public beta to run always-on background workers like Temporal and Celery without manual cluster management.
3. 15:21 - 24:30

   The Agent Stack: SDKs, AI Gateway, Sandboxes, and Drives

   Building production agents requires unified coordination across models, tooling, durability, and secure execution. The Eve open-source framework couples with Vercel Connect to consume events directly from MCP servers without custom authentication boilerplate. In addition to processing 4 trillion tokens daily through model routers that cut integration overhead, secure execution occurs inside 6 million daily microVM sandboxes. Vercel Drives resolve the problem of ephemeral compute by decoupling file storage from VM lifetimes, letting subsequent tasks mount persistent file systems, reuse node modules, and carry forward project state.

   • The AI SDK processes over 30 million weekly downloads, paired with the Workflow SDK for step-level state durability.
   • AI Gateway routes across hundreds of models with per-key spend limits, failover protection, and zero data retention.
   • TokenPass provides $80 of token credits for $30 per month on selected models with strict privacy constraints.
   • Vercel Drives offer persistent workspaces that mount to isolated Firecracker microVM sandboxes across tasks.
4. 24:31 - 36:23

   Enterprise Governance, Bring Your Own Cloud, and Internal Agents

   The rise of non-engineering citizen developers creating custom dashboards and tools introduces severe data leak and shadow IT risks. Centralized control planes enforce private-by-default access, task-scoped short-lived credentials for external systems, and hard budget caps per project. Enterprise customers can also isolate compute workloads within their own AWS perimeters to satisfy strict compliance needs. This combination of accessible tooling and strict governance enabled Klaviyo to deploy 356 apps across 512 internal builders in two weeks, while Vercel automated nearly its entire tier of customer support.

   • Identity-provider integration automatically revokes deployment and app privileges when staff leave Okta or Entra.
   • Bring Your Own Cloud on AWS is generally available, letting organizations run Vercel inside their existing AWS infrastructure.
   • Centralized spend limits and security audits prevent rogue token consumption and exposed credentials across internal tools.
   • Vertex, an internal support agent built by support staff and hardened by engineering, resolves 91% of Vercel support cases autonomously.
5. 36:24 - 39:23

   Scale Milestones and the Future of Computing

   Infrastructure is converging around platforms that natively support both human builders and fleets of autonomous coding agents. Every tier of modern application delivery—from frontend user interfaces and Python APIs to background workers and stateful model orchestration—now deploys under a unified control model. Rapid cycles of agentic software deployment allow development teams to deliver projects orders of magnitude faster than through manual coding workflows.

   • Vercel serves 4 trillion tokens daily, 60 million weekly Next.js downloads, and over 100 new sandboxes every second.
   • The platform unifies frontends, Python APIs, arbitrary container runtimes, and always-on daemons under single-click deployments.
   • More software will deploy over the next year than across the prior history of computing.