As far as I can remember, around 1995 or 1996, I was interested in the migration or transfer of software objects from the perspective of object-oriented development. In fact, I think it was IBM, in partnership with Apple, that created Taligent. The appeal was that software module could go wherever it was possible and execute the work. As far as I remember, it was based on micro-kernel, and the modules on top of it didn’t necessarily have to be on the same machine, so I think that was the image of the implementation design. There was a company headquater, or rather a building with name hidden, in Milpitas. Through some connection, I was asked to come, so I drove all the way from RT101 to the very end, and at the dead end, I think it was to the right at the Chinese restaurant I used to frequent, and it was right there. That’s where they were working. I think I still have some of the details somewhere. Around that time, I think it arose spontaneously, but the prototype of Java was born from the challenge of porting, starting, and integrating software to home appliances. I had some close friends in that field, and eventually I ended up helping to finalize the Java language specification for Addision book. The book includes an acknowledgment to me. All of these things were interconnected. While Symbolics, which was financially precarious, had a somewhat practical AI system running, the question arose of how to carry it over into the rapidly expanding internet boom of the time, and we began searching for an exit strategy.
And then there’s the “agent.” I don’t think there’s a definitive definition of this term yet. All of the above dreamed of agent-based development, albeit with different dreams. Wanting to secure a solid path, I translated “An Introduction to Intelligent Agents using Java” during my sabbatical at the time. It was a small-scale conceptual implementation, but at least it provided a steady path to application and development. This was in 2002. https://prof-ida.com/Books/CIAJ/index.html
Looking at IBM’s AI agent development now, although I haven’t looked at it precisely yet, it’s not impossible that it’s an extension of this. On the other hands, many AI agent configurations seem to center around the ReACT model, and this round-robin approach is reliable but primitive. Therefore, it’s easier to implement.
This is the historical flow from 30 years ago to the present day, something I haven’t written about at all until last week from now. Of course, it’s from the small window of the world as seen by the author.