Up until now, computers have required us to create new interfaces for them to understand us. A graphical user interface (GUI) is about creating graphics, buttons and other bridges into our world so we can instruct the computer on what we want to do. Prior to that, you had a command-line interface doing the same thing with text and commands. Even code itself is an interface - a way for us to instruct the machines.
All of this is a easier than working with low-level machine code or flipping bits directly, sure. And now we have an even 'easier' interface: generative AI. With large language models and their understanding of natural language, we now have the machines at the point where they can interface with us the way we interface with each other.
I think this leads to an interesting forced shift in perspective.
We are no longer having to meet machines in the world of binary and 1's and 0's. Instead, we have software meeting us in the world of probabilities, uncertainties and the imprecision of language. But as that happens, we need to adjust our expectations. Because if we are used to providing specific instructions and having calculated, specific, deterministic results - that's quite a different paradigm to asking in terms that can be misunderstood and misinterpreted and which generate results that are open to interpretation, creativity and hallucination. Not a problem if that's what we're expecting, but if we're still acting like the 'old world', then we have an expectation mismatch.
Also, there are certain restrictions from the old world that no longer apply. The change in interface means a real change in availabilty of interfaces too. If you have something like modern agentic capabilities that can use a web browser like a human, or examples where an AI mechanism can call and sound like a human in order to make a restaurant booking, then there's no need for the restaurant to create an API for programmatic access. The agentic capability can just use the interfaces designed for the humans, including the human staff.
Suddenly, suppliers are not constrained by not having the time or the technology to create the interface for machines to use. The flip-side is they now need to deal with a greatly increased amount of potential 'customers' (or agents thereof).
APIs were the first real attempt at machine-to-machine communication - software talking to software - but it wasn't software talking to humans. We've gone from humans talking to computers, to computers talking to computers, and now computers are talking to humans to get things done.