I was doing a mental exercise these days, trying to anticipate what will happen with generative AI technology in the next chapters.
My final thesis in Computer Science was about the semantic web. The era of intelligent agents through an internet structure more prepared for computers to understand the world. And not only syntactic, for humans to understand, as it is today. And I noticed how, in a way, we did follow the path of creating a semantic structure for autonomous agents to understand, but that structure was built through what I call brute force. With a lot of computational power, we took the vast digitized content of the internet, including curated posts and all kinds of information available there at massive scale, and with LLM technology we got a source of truth for agents to navigate. So the old concepts that agents would use ontologies and semantic representation technologies, such as RDF and OWL, were not used, in this first public phase, in the way imagined back then by Tim Berners-Lee, the creator of the concepts behind the WWW.
But now we are in a new phase. The truly agentic phase, and I believe it makes sense to bring those technologies back into our local solutions, as a kind of RAG for LLMs.
During my experience in IT, I often integrated principles such as Mobile First, API First, Serverless First. All of them worked as a guiding constraint so the project could be born with a clear path of priority. And now I am thinking about Agent First Development.
In my current agent first workflow, I have the "root agent". It is like a master agent that keeps the immutable guidelines for all the other agents. I started to stop thinking about software as SaaS. And I imagine software will become so easy to create that more and more people will create it on demand.
I am already doing this. For example, for financial investments. Instead of hiring a SaaS for investment portfolio management, I created my own. Customized for my needs. And imagine this in a pair of smart glasses, where you say what you need and the software is created on the home screen of your device in real time.
Thinking now about my current work workflow, I no longer have SaaS projects. I have what I called initiatives. And each initiative is guarded by a main agent, which can instantiate more agents to complete tasks. And because this agent was born for the purpose of the initiative, it learns more and more about that domain and area of knowledge and becomes a specialist in that initiative, learning more and more through "experience". All of this guarded by good methodologies and practices that we humans used in our processes, but now delegating to agents the work of following those practices and processes.