What is an agent and how can we design useful systems using agents?
That's a question I spent a few years of my life considering, leading to a PhD thesis, a few academic papers and a book. It's also a question I had parked, for almost 15 years, because although interesting it did not solve any pressing problems. While we could hypothesise about agent systems, the AI technologies required to build them at scale where not there. Like much of AI at the time it "showed promise" but required "further work". It was the very definition of an "academic exercise".
These days, however, that question is back. More importantly, it is within reach. It's an issue I see people and organisations grapple with as they are trying to figure out what is and what is not an agent, what they can do with them, whether and how to build them and how will they charge for them.
This was the impetus behind starting AgentsDecoded. It is a subject that, despite the fact I've already spent years on, I am still excited by. As an entrepreneur and technologist with an academic background in the space it feels like I could offer some useful thought and analysis.
We will be doing two things at AgentsDecoded:
- First, we are going to ground our understanding of AI Agents in a simple, universal framework. This will allow us to analyse anything that presents itself as an AI agent, better understand it and compare it to other such things.
- Secondly, we are going to track news and analyse developments from a business, technology and academic perspective.
The focus will be on as clear a signal as possible and as applicable advice as possible. There is a lot of hype in this space - as with any nascent field - and it is easy to get carried away with enthusiasm. We will work hard to bring as much clarity as possible to the subject.
We've defined a goal and explained some of our motivations. Time to act.