OpenAI Operator and how we will consume services
With the release of OpenAI's Operator I'm seeing people saying "oh we are going to need to rebuild websites for Operator" ...really?
if you need to rebuild, it completely misses the point of having an agent that knows how to navigate the web.
whatever you do it won't ever be better for an agent than a simple API spec.
you are going to end up with a UI that is bad for users and worse to agents than any API would ever be.
Operator seems very interesting and I do think agents using browsers is a piece of the puzzle but I'm skeptic this is how agents will work in the future when they are taking simple voice commands and executing complex tasks in the background by integrating a bunch of services.
Provide a `openapi.yaml` file of your API and you have an agent that is much more efficient in accomplishing any task than clicking through a browser will ever be, with the benefit that it can present results in any format it's best for you to consume.
While browser navigation will always be necessary for certain things, most of the services we consume in the day to day already happens through an app - which does things through an API.
In practice, if services want user to provide the best possible experience to users it's in their interest that agents can navigate their content and execute tasks in the most efficient way. So what I believe is that we will end up with a kind of app store which instead of UIs to interact with APIs, agents will just have the description of the APIs. They will execute the tasks through these APIs and generate an UI to display the result to users - or just speak to us directly.
Please consider supporting me by subscribing to this newsletter and buying my book Strategic Software Engineering on Amazon.