In 2023 - I started to understand how important memory is to the system, and the intelligence is a storage problem as much as it is a processing problem.
In 2024 - I started to realize that AGI is the wrong term all together and that intelligence can't be solved with scale alone.
In 2025 - I started to understand just how many components needs to work together to build a real agent. We started calling them agents because they have many different moving components nothing else described them. RAG became obsolete as it was too rigid to keep up. Not only we needed an LLM, memory & orchestrator, but also perception & RAI.
Now in 2026 I am starting to understand just how complex intelligence is, which I would have understood sooner if I took more biology courses in uni. I remember an episode of "love, death & robots" from 2022, in which a human scientist gets invited into an alien nest/entity that has grown to an unimaginable scale. He goes inside to explore/communicate, finds a swarm of workers/warriors but no centralized intelligence. After being captured by the warriors, a human head evolves infront of his face in seconds to communicate. A brain that is more intelligent than anything he ever encountered in the galaxy, surprisingly that intelligence was humble enough to understand it's place in the swarm because intelligence is just a component of the system. It spawns into existence when a new lifeform gets close and goes vanishes afterwards.
I am reading more now about the brain, everything we built by now around LLMs is still inferior to the Broca's area, one tiny part of the brain that is responsible for language & speech. Its not just about building processors, the harness is much much harder. The human brain surprisingly has a very minimalist UI, each component is able to grow more or shrink bases on what is required in real time. And the hardware changes following the changes that happen in its software. Psychology changes change brain structure, and brain structure.
What we are doing on the other hand is building rigid harnesses around LLM instead. Its like if the brain had only 1 region that is truly adaptable where all the other areas are uniform across species & individuals.