If you follow the news about AI, it quickly becomes clear that it’s here to stay. But if you take a closer look, you soon realize that it essentially means turning off your own thinking. No matter which blog or podcast I follow, the result is (almost) always the same: everything is getting easier and faster. But that’s only partly true.
Of course, creating graphics and texts is much faster and, in terms of content, probably more comprehensive than if it had been done manually. What’s missing, however, is the personal touch of the author. I subscribe to a newsletter precisely because I want the author’s personal perspective—not a quickly cobbled-together text or a bullet-point list of news. I can find that anywhere.
And now (almost) everywhere, the latest hot thing—agents—is being hyped. But what exactly is that?
Essentially, it’s about giving an “agent” (or personal full-time assistant) a task that it then works through independently. No matter what can be done online, the agent can do it for me—from booking appointments to shopping, and just about anything else. In theory, I think that’s fine, but it comes with significant challenges. And that’s what this is really about.
Every company that wants to remain relevant in the future must provide an API that enables access to content and facilitates interactions. In other words, any company that wants to sell or book something online needs an API so that agents can use it.
Mere interaction in the form of conversation will not be enough in the age of agents. Simply finding content—whether images, products, or events—won’t suffice. It must also be possible to purchase products, book events, vacations, or services directly through agents. And that will only work if a suitable interface is available. In today’s context, this means that every company selling online needs an API that enables both discovery and transactions.
In the future, this API will become a standard item in every website’s main navigation—whether in the header or footer doesn’t matter, but it will appear right below “Legal” or “Imprint” to ensure both users and agents can easily find the API specification.
Additionally, this API will be listed in a standardized way, reflecting its growing importance. Companies will establish a convention to host their API under api.domain.TLD, where both the specification and usage details will be readily available.
Once an API is in place to provide access to an offer—whether as a list or in a conversational format—the agent ultimately reaches the point of placing an order. And this is where the question of security arises. Is it enough to simply pass a payment method to the API, such as a credit card including the CVV number? Or is a login—and therefore a customer account—required? How much do we trust smaller providers to keep our data secure?
If an account is needed, managing all login credentials could quickly become overwhelming. Should the agent be connected to a password manager instead? That would mean the agent itself would need an API to securely access passwords and payment details when needed.
In my view, these are the two most critical requirements—both for providers and users of agents.
To truly usher in the age of agents, I can imagine the following steps:
Currently, AIs are relatively well trained to generate coherent language in a human-like manner. However, their "knowledge" is limited to their last training cutoff. But in the context of agents, this isn't necessarily a problem—they don’t need more knowledge, just the ability to understand what needs to be done. For that, today’s models are already sufficient. The only missing piece is real-time data.
This can be solved by providing the AI with an API specification as context, allowing it to retrieve up-to-date information directly from providers. Google, among others, is already implementing this with the Search API, along with various other providers.
By embedding the API spec and access credentials in AI interactions, it’s already possible to work seamlessly with different providers purely via interfaces. Whether it’s file uploads, image analysis, or even online research—the AI can interpret the API documentation and act accordingly.
Just as the term "prompt engineer" was the hot new job title about two years ago, it’s already becoming obsolete. AI models have advanced to the point where we no longer need prompt engineers but rather context engineers. For high-quality results, the context of a query is now more important than the precise formulation of a prompt.
But even this role will soon fade, as every halfway decent website will offer an API accessible via api.domain.TLD, making manual context engineering unnecessary.
Once a search yields relevant results, the agent’s next (automated) step will be to load the API and interact with it according to its specification. While in Phase 1, we still had to explicitly tell agents what the API spec looks like or where to find it, in Phase 2, this will become the default—no manual intervention required.
By Phase 3, we will have fully embraced agents. I can already picture myself telling an agent to buy a gift for my best friend and have it sent immediately. The rest happens in the background within minutes: the gift is searched for, found, and ordered, with payment automatically deducted from the stored credit card. Whether it was a good choice? We’ll find out when it arrives.
And while writing this, I realize—once again—this is yet another task that no longer requires my brain. Years ago, I stopped memorizing phone numbers. My phone and its backup made that obsolete. Why burden my brain with such trivialities? The same applies to AI-generated content—whether images, audio, or text. Today, it’s already enough to write a halfway decent prompt and provide the AI with some pre-existing text as context. The newsletter or cover letter is then created almost like magic.
So why am I still typing this myself? Couldn’t I have just fed my key thoughts into an AI and let it generate the text? Well, so be it… After Phase 3 of the agent revolution, I won’t be worrying about such things anymore—I’ll just let the agent handle it. Let’s hope it all goes well.
The age of agents is coming—and those without an API will be left behind.
At diconium, we have years of experience in developing microservice architectures, APIs, and interfaces. We can help you get your shop or offering ready for the era beyond tapping, clicking, and swiping.