Choose the Layer You Can Live With

A platform earns its fee when it removes work somebody would otherwise have to do. It becomes expensive in a different way when its limits dictate the customer’s process or make an ordinary change difficult.

AI assistance gives experienced builders more room to question that trade. Infrastructure code and workflow logic can be easier to produce and examine, so a lower-level service may be a practical choice where it once demanded too much effort. That does not make operating it free.

Count the work on both sides

Managed tools can provide deployment, monitoring, upgrades, and support. Composing cloud services can give a team more control over behavior and clearer access to its own code. Either approach can also leave the customer with dependencies they do not understand.

The useful comparison includes who will look after the result, how they will diagnose a failure, and what happens when the team wants to change direction. The few hours saved on the first build should not settle those questions.

Be precise about ownership

Running a capability in a customer’s account can give them control of access, billing, configuration, and source. It does not mean they own the cloud provider or can move to another one without work. We should make that boundary clear.

A state machine or other explicit model can help people inspect a workflow, but the model remains an account of the work. The people doing it will find exceptions that the diagram missed.

We prefer understandable tools that the customer can live with. Sometimes that means composing simple services. Sometimes it means buying a platform and spending our time on the customer’s problem. Getting closer to the metal is useful when it serves that choice, not as a badge of seriousness.

Original illustration generated with AI for this belief.