Most business processes have a step in the middle where somebody has to look at something, work out what it means, and decide what happens next. Software has always been able to do the steps either side of it. Language models can do that middle step now, which means the whole process can run without a person sitting in it. Building those is the work, and making them dependable enough to leave running is the hard part.

The product your customers use, and everything it needs to keep working. There is more to it than the application: content somebody has to be able to change, money moving through it, and the place it all runs. They are usually built at different times by different people, and getting them to work together is most of the job.

Sometimes a team is building well and simply needs more senior hands. Sometimes there is nobody whose job it is to decide what gets built at all. We do both, and they are different jobs: one adds capacity to a plan somebody else owns, the other means owning the plan.

Every system has to know who someone is before it can decide what they are allowed to do. The first half is usually bought and the second is usually built, and both stop fitting as a company grows: providers get changed, big customers arrive expecting to sign in with their own, and a short list of roles stops describing who should see what.