The promise of vibe coding was that full-scale applications could be spun up from just an idea, powered by various AI systems. But it turns out writing the code is just the first step in the process. Vibe coders are already running into the standard headaches of maintaining and updating a software product.
Luckily for them, a new crop of startups is arriving to fill those gaps. The platform engineering startup Shuttle recently announced seed funding to handle the infrastructure problems that begin where other AI coding products leave off.
Shuttle will take code produced by a vibe-coding system and assess the best way to deploy it. It presents the user with a sensible infrastructure package along with a price tag. Once the user agrees, Shuttle can arrange payment and deploy the software directly to the cloud provider with minimal friction.
It has been a long road for Shuttle, which launched as part of a Y Combinator class several years ago. Since that time, it has become one of the most popular systems for deploying Rust applications, drawing in thousands of developers across many deployments with a fast zero-config approach. With new funding, the company plans to expand that to every programming language and AI coding system.
As the CEO describes it, agentic AI systems have made the barriers between different programming systems much easier to cross. This means a system like Shuttle can be deployed in all of them at once. AI is wiping away the borders between different language ecosystems, making it a perfect time for the company to scale up after years in back-end development.
In practical terms, this means building an agentic interface for platform management. Users can provision a database or purchase cloud hosting using the same natural language prompts that they used to vibe-code their application. On the back end, it also means building interconnections with cloud providers and coding systems so that the agents have all the context they need.
The company has created a specification that works as an intermediate layer between what humans are able to review and what AI understands. Spec-driven development is becoming the go-to way of doing things, and there is no reason that should not be the case for infrastructure as well.