There was one aspect missing: you can use a stack layer that is "rented", but when you have a clear interface you can swap the layer to someone elses stack. When the interface is not clean and abstracted, you will run into trouble. This was always the case in software architecture and this will be even a much bigger problem when people do not understand the layers of an agentic AI stack. It's a design choice. You cannot and should not build everything by yourself, but you must right from the start understand how to design a system with interfaces that allow you to change vendors or suppliers
I agree with your point. However, clean abstractions require upfront investment that the market actively punishes in early stages. Speed gets rewarded and tight integration ships faster. The founding team that stops to design a proper abstraction layer over their LLM provider while a competitor is moving is making a bet on future resilience over current velocity. Most investors won't value that bet. Most early metrics won't reflect it.
It's a constant trade-off. And there is not that one solution. There are different layers here. Individuals and also companies won't be able to solve some of those. It's about incentive structures and about alternatives. Some of those need to be dealt with on political level. But that's easy to say as that's pushing responsibility outward in a layer that that is notoriously slow in executing.
It could even be a mixed calculation. Go for speed in the beginning, later refactoring or rebuilding in parallel on a different stack with a different design.
On Your Terms - On the freedom to depend without surrendering agency.
--> I really like it!
That will be a great topic and appeal to many entrepreneurs to think about.
Happy to collaborate on an impact angle on it!
Yes. Glad it resonates. And very happy to work together to add that impact angle.
There was one aspect missing: you can use a stack layer that is "rented", but when you have a clear interface you can swap the layer to someone elses stack. When the interface is not clean and abstracted, you will run into trouble. This was always the case in software architecture and this will be even a much bigger problem when people do not understand the layers of an agentic AI stack. It's a design choice. You cannot and should not build everything by yourself, but you must right from the start understand how to design a system with interfaces that allow you to change vendors or suppliers
I agree with your point. However, clean abstractions require upfront investment that the market actively punishes in early stages. Speed gets rewarded and tight integration ships faster. The founding team that stops to design a proper abstraction layer over their LLM provider while a competitor is moving is making a bet on future resilience over current velocity. Most investors won't value that bet. Most early metrics won't reflect it.
It's a constant trade-off. And there is not that one solution. There are different layers here. Individuals and also companies won't be able to solve some of those. It's about incentive structures and about alternatives. Some of those need to be dealt with on political level. But that's easy to say as that's pushing responsibility outward in a layer that that is notoriously slow in executing.
It could even be a mixed calculation. Go for speed in the beginning, later refactoring or rebuilding in parallel on a different stack with a different design.
Yes. This requires conscious choices. It's a constant battle between the urgent and the important. Just like technical debt.