DEvops
CTO claims cost premium for using AWS indirectly via Vercel is mitigated by more efficient use of computing resources
Vercel introduced an open-source agent framework called eve at the Ship event in London this week, along with other new features including Passport, an attempt to bring AI-powered employee apps under enterprise control.
Agents, especially custom agents, are currently dominating AI conversations. Agent frameworks already exist that simplify coding, but eve has some notable features.
The coding languages are TypeScript and Markdown, and the agent is a directory containing files that define instructions and skills, model providers, tools, authentication, channels, and schedules. By default, agents are sandboxed on isolated VMs. The framework also includes simple testing tools to run the agent and evaluate the results. The code is on GitHub under the Apache 2.0 license.
While there are many existing agent frameworks, Vercel CTO Malte Ubl said eve’s hallmark is its simplicity, allowing users to take a “fill in the blanks” approach.
“The agent lifecycle is completely orchestrated by the framework, and the developer or builder has to put things in the right place, but then everything works like a charm,” Ubl said.
“This is a system where you don’t have to understand bits and pieces of what a sandbox is and how to compress context windows. All of these things are very complex. You don’t need to understand any of that.”
Agents built with eve are deployed to Vercel by default using the same commands that work in web applications. versel deployment. That said, the company says it is not platform bound.
“We’re working hard to make it work everywhere,” Ubl said. However, early users have already raised issues with the Vercel login being required even when configured to use a different model provider. It’s still in the early stages so it may be a bug. LLM and sandbox providers are configurable. The Eve project is also run locally using: npx eve developer.
Which LLM does Eve use? “It can connect to any model that the AI SDK connects to, so any model,” Ubl said, noting that the AI SDK is the Vercel SDK. You also have the option of using Vercel’s AI Gateway. It has a single endpoint for multiple model providers and can improve reliability by switching to another model if one model fails.
The company also previewed Enterprise Apps and Agents, which consists of four components. versel connect Replace static secret credentials with short-lived tokens accessed via OAuth or API. versel passport uses OpenID Connect to put all your applications and AI agents on a team behind an identity provider like Okta or Microsoft Entra. Enterprise Managed User uses directory synchronization to enable Vercel in your team to be managed by your organization’s identity system. Finally, Deploy your own cloud (BYOC) allows organizations to use Vercel’s platform running on customer-provisioned AWS infrastructure.
According to Vercel, Passport was a highly requested feature because many employees create applications that are hosted by Vercel but outside of the organization’s control. A common scenario is that employees build applications with AI assistance, and the AI agent uses the Next.js React-based framework and Vercel hosting by default. This is a type of shadow IT (or shadow AI) in which staff use company data to create vibe-coded applications that are outside of the organization’s IT policies and control.
Vercel itself is an AWS customer, so its platform should work well with BYOC, but there are some tradeoffs, Ubl said. One is “We don’t allow your compute to assume AWS roles… If you have a deep understanding of AWS IM. [Identity Management] “But we always issue an OIDC token for every compute call, so we can use that to set AWS policies.” Second, with BYOC, “we become a managed vendor,” Ubl said. This means giving Vercel access to that part of the customer’s AWS infrastructure.
All Vercel deployments are immutable. This means that “every time you push to Git, you’re getting new infrastructure from scratch,” Ubl said. He thinks this is ideal for AI agents. Other aspects of the platform are also optimized for agents. “We try to be closer to agency work,” he said.
A common criticism of Vercel is that because it runs on AWS, when you use Vercel you end up paying lower hosting fees than if you purchased it directly. According to Ubl, that premium is mitigated by Vercel’s efficient use of these resources, “especially at smaller scales, especially compared to Lambda, which is an AWS serverless platform.” Vercel announced last year that it had reduced Lambda costs by up to 95% by reusing idle instances.
Ubl argued that AWS customers need “a usage rate of 35% or higher to match Vercel’s price.”
Another competitor of Vercel is Cloudflare. Cloudflare, unlike Vercel, has an efficient serverless platform that is hosted in its own datacenters and uses Workers, a feature of the V8 JavaScript engine used in Google Chrome and the open source Chromium project, based on V8 Isolate.
Ubl said that while Cloudflare Workers is unique to Cloudflare, Vercel is “more of a regular platform, we don’t run a custom runtime that we write ourselves, we just run Node.js, Python, PHP, it runs on a VM (virtual machine)…We offer standard PostgreSQL, VPC peering, AWS, S3, it’s not custom-built.”
This is a bit of a war of words. Steve Faulkner, Cloudflare’s director of engineering, said in February that the Vercel-sponsored Next.js tool is “completely bespoke.” Things have improved since then, with Next.js 16.2 stabilizing the adapter API so other providers don’t have to reverse engineer the build output, but adapters for AWS and Cloudflare are still in development and expected to be completed by the end of 2026. ®
