AI Agents or “Agentic AI”

Recently, “the age of Agentic AI” was announced to be upon us. The announcement provided a look into the near future. Agents in the form of autonomous actors, leveraging potentially many different conversational AI and LLMs to accomplish common tasks, are here to help us. The perspective is of course driven by the incentives for the large investors in the field of AI development. So let’s take a look on who this is going to help and what providers of actual value adding services can do to get their piece of the pie.

“Where’s the money, Lebowski?”

It is obvious that most spending from consumers will go to parties that provide the access and orchestration of conversational AI and Agents. These are the Apple, Microsoft, and OpenAIs of this world. They have exclusive positions to ring fence access and act as gatekeeper to the technology. But, there’s more to this. Because this is not the only game in town.

Do the thing

I was contemplating on a quote by Chris Williamson of the Modern Wisdom podcast. He said something along the lines of:

“Talking about the thing is not doing the thing”
“Planning to do the thing is not doing the thing”
“Posting on social media about the thing is not doing the thing”
…

You can guess where this is going.
Full post here

Also, great podcast!

The gist of the quote is very close to what agentic AI brings us. A lot of talking about the thing, planning things, scheduling things, but not really the capability to actually DO THE THING!. And precisely this is where providers of actual value adding service have an opportunity.

Do it well

The power service providers have, is that they can make a difference. By applying local knowledge, unique perspectives and opportunities formed on- and off the internet actual value add is generated. Providing access to these these services will always be valuable, potentially scarce and unique. If your service or product has one or more of these qualities, it will have enduring value.

Build an API for it

To be able to provide such a service over the internet and “plug in” to the Agent landscape, a well-defined API is key. Providing an API for a conversation requires a different approach in modeling the business domain then we were used to. One needs to think in decision making, communicating instructions and understanding how your business service can be consumed. We will see significant improvements and amounts of experiments in this area over the next years.

Monetize it

Monetize it may sound a bit awkward, but it is key to ensure you get paid for providing your services. Do not wait until you have finalized your service idea, but really include this as part of the conversation you want to provide. If you patch this on as an afterthought, chances will be high you’ll miss the opportunity to be part of the conversation (phun intended). So, think on how to monetize a conversation and how you intend to get paid, by whom and when.

Payments need to be PART of the API

If getting paid requires human intervention, you are not going to benefit fully from service consumption by AI agents. So, think about providing payment instructions or provide fully integrated ways of exchanging value. Also think about your audience. Are you providing business-to-business services or are you providing business-to-consumer services? The nature of the interface can change depending on the way you sell(monetize) your services.

Business to business

Although an agents represents a single actor, an actor is employed by a company. It may make sense to provide pay-as-you-go and pay-per-use options, but to generate recurring revenue, possibly a capped rate of use, billed per Month may provide a better options. Think about this really thorough, because changing the payment mechanism afterwards will disrupt actors.

Business to consumer

For business to consumer services a pay-as-you-go option makes sense. However other options are available as well. One can opt for micro-transactions that don’t require an intermediary bank to settle a payment. A way to for example think about this is using the Bitcoin Lightning network. One can stream payments for maintaining the conversation, think of this as direct payment for each generated token.

I personally think the frame of business-to-consumer best represents how to think about the monetization of a service provided through agent conversations. This is scalable, has value exchange built in, directly tied to the interaction duration and value offered by the service.

Banks, regulations and law

It will take a while for banks to catch up to these requirements and seamlessly support micro-transactions in fractions of the money we’re using to support these conversational use cases. I think we’ll see big tech push for regulations change here as well to enable new business.

Businesses need to operate within a legal framework, so be careful with embarking on a journey where your services are provided cross geographies without respecting existing trade regulations and law.

From the agentic perspective, it’ll be interesting to see how countries will think about the form of employment AI agents provide as part of a workforce. The next few years will definitely show progress on how this is regarded and how societies should benefit.

Protect your unique value

Whatever you do, ensure you properly protect your unique value. A part of basic protections, think about protecting your services against unwanted reselling, fraud, arbitration, against over-use by applying rate limits, and by protecting the thing you do well as trade secrets. This ensures you can stay in the game.

Wrapping up

It’ll be interesting to see how agents will become part of our world and to see how they can and will provide value. For providers of uniques services it will be hard to compete in the short term using the traditional APIs. Especially since in the short term the tools to design and build conversational APIs for the Agentic AI era are not here yet. As software developers this is exciting, because this opens up a new area of investment for companies that have legacy APIs.

So, there’s the money Lebowski!