Explain how new technologies, like LLMs and ChatGPT, may revolutionize Application Programming Interfaces (APIs) in the future. Provide an example of how these APIs might work.
As a kid, I loved watching the Jetsons. As I grew older and ended up writing software and creating & implementing architectures for systems, I thought I would never see the day when a Jetsons-like kitchen would be feasible.
GPT (ChatGPT, etc.) has changed my mind.

Contractual Obligations
As an architect, you kind of need to think like a lawyer. You define contracts (APIs) between systems or components. You monitor those APIs for abuse or misuse.

I would use Swagger editor to define those contracts, and I would write the APIs to match interactions implied in mock-ups. I would vet the design with the team, I would walk through scenarios to see if the API would fit the use case and would not introduce performance bottlenecks, and I would simulate and automate as much as I could along the journey.

I was very careful to get it right the first time. It was never perfect the first time, of course. So, I also had to consider how the API could evolve while minimizing costs/impact to the rest of the system. The API, once deployed, would become something I would guard, like an honest politician guards the #usconstitution. The team had to be careful with any change that impacted the API.

API Mesh... More like API mess!
APIs rely on APIs which rely on more APIs and libraries. Once a system gets large enough, that complexity becomes a bottleneck. Small changes can have a huge ripple effect. Some approaches (like federated #GraphQL) help manage this complexity nicely. But it still involves an API under which there can exist a lot of fragility. #Agile or not, developing a complicated web of APIs is complex, requires specialized skills, and is full of unforeseen challenges.

The Kitchen Conversation, ChatGPT-style
The Jetsons' kitchen involves a number of systems, each composed of various parts, sensors, hardware, inputs and outputs. Considering the complexity of interactions between independent systems in a kitchen, the fact that the Jetsons' kitchen components would come from various vendors or manufacturers, traditional API design would not be able to fulfill the vision. Not practical, at all.

Now, let's think about APIs and the Jetsons' kitchen, #chatgpt -style.

Let's keep it simple: pizza. You (or your robot) make a pizza from scratch.

The oven asks the refrigerator: "I see we have a pizza. Can you tell me if this is low-moisture mozzarella?"

Refrigerator: "That's not even mozzarella. This pizza must have Muenster cheese as that is all I had in me. And the human removed some of the contents 30 minutes ago, so the probability of Muenster cheese on that pizza is quite high."

A sensor from the kitchen who has eyes across the room chimes in at this point. Sensor: "Beware that the crust of this pizza is made of cauliflower. I agree with the video sensor's assessment regarding the type of cheese. No new cheese entered the kitchen recently."

Oven: "Anything else I should know?"

Sensor: "The temperature of the dough is 47F. This is cooler than the expert chef consensus of using room-temperature dough."

Oven: "Great. I know what to do. Since the dough is cool, I will start cooking with a temperature a little lower than the recipe specifies. Then I will increase the temp of bottom element so as to provide a nice crisp crust. I will keep a close eye on moisture and may use convection to mitigate any undesirable levels of moisture."

Where is the API that contractually binds the independent systems? APIs are not gone. For instance, the brain of the Oven will have myriad APIs (e.g. for interacting with temperature controllers and sensors, etc.). The key here is that the interface between systems is no longer a fixed, rigid API. No method signatures, no paths, no payload specifications. Instead, a conversation relating to senses and capabilities.

What "brain"?
Kence Anderson's book Designing Autonomous AI describes nicely how a complex system can be decomposed into a number of independent "brains." Each brain, in his book, can be independently developed and refined, with human guidance married with #datascience, #machinelearning and #dataengineering.

What's Cookin'?
I am excited for what the future holds. How architects rethink complex system design with regards to APIs is, in my opinion, something that is inevitable. And, as an amateur pizza maker, my taste buds are watering for the day I get some GPT help to make an even better #detroitpizza.