Workers for Platforms

If your product lets customers run their own code — plugins, functions, integrations, custom logic — Workers for Platforms is the only managed offering that takes the problem seriously. Dispatch namespaces hold unlimited user Workers, a dispatch Worker routes to them, and custom limits plus outbound Workers let you sandbox what tenant code can do and see. The hard part — actual isolation with per-tenant resource limits — is the platform’s problem, and the alternative is building your own multi-tenant isolation on containers, which is a full engineering team’s job to do safely.

The gaps are what pull the score down. Tenant Workers don’t get the full platform: no Durable Objects, which cuts off a whole class of stateful use-cases the main platform sells hard. There’s no preview or staging system for dispatched Workers — tenants deploy blind compared to the first-party experience. And Workers observability barely reaches into dispatch namespaces: no proper logs, traces, or per-tenant error aggregation, so surfacing “your integration is failing and here’s why” to customers means building that pipeline yourself.

Nothing else in the market really competes with the isolation model, which is why this isn’t lower. But the difference between “unique primitive” and “complete platform” is exactly the difference between what it is and what it charges like.