Vercel, Cloudflare, DigitalOcean Race to Build Infrastructure for AI Agents

Vercel, Cloudflare, and DigitalOcean each released new infrastructure tools and guidance this week aimed at the growing wave of autonomous AI agents, signaling a shift in how cloud platforms are adapting to agent-driven workloads. Vercel debuted an open-source agent framework called eve at its Ship event in London, while Cloudflare rolled out temporary accounts for its Workers platform. DigitalOcean published technical guides on server-side tool architecture and LLM compression.

Vercel's eve framework, licensed under Apache 2.0, lets developers build agents using TypeScript and Markdown. An agent is defined as a directory with files for instructions, skills, model provider, tools, authentication, channels, and schedule. Agents are sandboxed on isolated VMs by default. Vercel CTO Malte Ubl said the framework aims for simplicity: "You don't have to understand every little bit about what sandboxes are and how to compact context windows."

Shadow AI and Enterprise Controls

Vercel also previewed Passport, a feature that uses OpenID Connect to put employee-built AI applications and agents behind an enterprise identity provider such as Okta or Microsoft Entra. The company said Passport was a highly requested feature because of the prevalence of shadow AI, where employees create vibe-coded applications using company data outside IT policy. Other enterprise features include Vercel Connect for short-lived tokens, Enterprise Managed Users for directory sync, and Bring Your Own Cloud (BYOC) for running Vercel on customer-provisioned AWS infrastructure.

• Vercel's eve framework is available on GitHub under Apache 2.0.
• Cloudflare's temporary accounts let any agent run wrangler deploy --temporary and get a live Worker in seconds.
• DigitalOcean's server-side tools tutorial covers architecture, latency trade-offs, and when to switch from client-side execution.
• DigitalOcean's LLM compression guide compares SparseGPT and Wanda pruning methods to reduce inference costs on GPU cloud infrastructure.
• All Vercel deployments are immutable, meaning every Git push creates new infrastructure from scratch, which Ubl called ideal for AI agents.

Infrastructure Implications and What Comes Next

The simultaneous releases highlight a broader industry challenge: existing cloud infrastructure was designed for human developers, not autonomous agents. Cloudflare's temporary accounts directly address this, as the company noted that "the moment an agent needs to deploy something, it slams face-first into a wall built for humans." Vercel's eve framework similarly abstracts away sandboxing, context window management, and model provider configuration. DigitalOcean's tutorials focus on the operational side, helping developers understand when server-side tools reduce latency and how model compression cuts costs. Ubl acknowledged that using Vercel on AWS carries a cost premium, but argued it is mitigated by more efficient compute resource use. The company is also working on making eve work outside its platform, though an early GitHub issue noted a requirement for a Vercel login even with a different model provider. As agents become more common, expect more platforms to follow with agent-native deployment models, identity controls, and cost optimization tools.