Since I started trying Anthropic’s Claude Fable 5 in July, I have become increasingly convinced that the inflection point in AI intelligence has already arrived. I think we are in the early stages of the AGI era.
Fable 5 is the model that brought me to that conclusion. I would call it a “singularity” in intelligence: its capabilities have crossed a threshold. What interests me most now is when those capabilities will become cheap enough to be widely available.
If I had to put a rough number on it, I would go as far as saying that it could already handle 80% of the work people do. That is my personal estimate, not a statistical finding. It expresses my view of its capabilities; it does not mean that 80% of jobs can be replaced outright today.
There is still a cost gap between a capability existing and being widely used.
Training a model like this takes substantial investment, and having it work continuously for more people also takes computing resources. Even when a capability exists, it is hard to use it everywhere it would be useful if it remains expensive. The level of adoption we see today may not fully reflect the capabilities AI has already reached.
Fable 5 has been made publicly available, and Anthropic’s launch announcement also describes its safety restrictions. Cost is certainly not the only factor affecting adoption, but I believe it remains a major obstacle to deploying capabilities like these at scale.
What interests me even more is that AI will participate in lowering its own costs.
As models become more capable, they may be able to contribute more deeply to chip design, software optimization, and infrastructure research and development. Better chips and infrastructure can, in turn, support more capable and cheaper models. This creates a feedback loop: AI helps improve the conditions on which the next generation of AI depends.
Following that direction, I expect AI chips to become more powerful and more energy-efficient, while the cost of a given level of capability continues to fall. If this feedback loop keeps accelerating, the pace of change could exceed what we would predict from past experience.
That is why I lean toward the view that we are already at the beginning of the AGI era. What matters most to that judgment is the capability threshold I believe has already been crossed.
The question I am watching now is when those capabilities will become cheap enough for widespread use. As costs fall, how many applications that still seem expensive and difficult to scale will become ordinary parts of daily life?
I think we have already seen the beginning. Over the next few years, lowering AI costs will be one of the key factors shaping how this change unfolds.