Terra now

Infrastructure changes the customer relationship

At Foody, the product sat close to an everyday transaction. At Terra, much of the work sits underneath other products. The output is not a single consumer experience but infrastructure that other teams depend on to build theirs.

A failure can appear far away from the code that caused it. A confusing interface can be repeated across many products. If the infrastructure fails, it becomes part of a customer's failure to their own users.

The test I use is simple to ask and difficult to satisfy: did the system help someone finish what they intended to do?

How do AI agents change infrastructure interfaces?

AI agents change who or what invokes a system. They do not remove the need for confirmation, observability and safe failure. Software has usually assumed a human developer translating intent into commands. Agents can plan, call tools and carry context across tasks, so the infrastructure now has to serve people and software acting for them.

It is easy to turn that shift into a prediction contest. The practical questions are more interesting. How is intent represented? Which actions require confirmation? What evidence survives after an agent makes a decision? How does a system fail in a way that a person can understand and repair?

The interface may become conversational or disappear into an agent loop. The system still has to be dependable, observable and safe.

Terra is not the conclusion to my earlier company stories. It is current work with different people, constraints and responsibilities.

A current checkpoint

The work is current, so this article is only a checkpoint: what I am exploring now, written clearly enough that the future can prove it wrong.