Agent-native
Nowadays an agent can build a complete product in an afternoon. Shipping it is another story. The product has to be set up on a platform, deployed and operated, and a traditional PaaS quickly becomes the bottleneck.
A platform built for humans has checkpoints you have to click through: a console setting, a form, an approval. That was fine when a team took weeks to build a product and each checkpoint marked a milestone worth celebrating. An agent builds one in hours, so the same checkpoints are now the slowest part of the job.
Agent-native means the agent does all the platform work, so nothing waits on you. On a cloud, it provisions all the services and connects them through the API, with no step that needs a console.
Then the cloud gets out of your way. You no longer need to know whether the code runs in a container, how the database is tuned, or how staging differs from production, because your agent makes those decisions now.
First, agents work best on Linux, the most standard and open platform there is. More has been written about it than about any other system, so an agent already knows it. A proprietary stack has to be explained to the agent at the start of every session, and even then the agent can only do what the stack allows.
Second, most of what a traditional PaaS sells is unnecessary. Its queue, search index, scheduler and metrics stack all wrap open-source software that anyone can install. Customers paid for the wrapper because installing the software took a week, and someone had to look after it from then on. An agent installs and runs the same software in minutes, so the wrapper is no longer worth paying for.
There is a benefit here that is easy to miss. Your agent sets up each service for your product alone, not for the average customer, so it is usually smaller and faster than the managed version, and sometimes it does something the managed one could not do at all.
We sell only the services an agent cannot build for itself. We do not sell observability, an API gateway, a serverless runtime or a queue, because an agent can build those.
An agent can write software. It cannot create hardware, an IP address, or a mail server that other mail servers trust. So we sell a machine you have root on, storage that grows and moves between machines, snapshots, an address, and an outbound mail relay. If we find something else an agent cannot get for itself, we will sell that too — LLM endpoints, DNS hosting and a second location are what we are looking at now.
Every service a platform offers costs money to keep running, and its customers pay for that. We run far fewer services, so our prices are lower.
Two things: the money and the kill switch. Only you can add credit or change the payment method. Only you can freeze or unfreeze the account. Everything else is your agent's job.