About agents, all the big players are shouting “AGENTS” and nobody seems to understand the difference that it means to the current tools that we have:

I assume that we are going to have two types of agents in the future: those built for enterprise and sensitive environments (healthcare, government, banks) that will require more human checks and will look clumpsy and too slow and too cautious and those built by start-ups and the open source community that are going to do a lot of stuff without asking and will look faster and more independent, and I bet the second type is going to be much faster to get to market and to have widespread adoption. I think that is what OpenAI, Anthropic, Meta and even Google are betting on for consumer use, but we will see soon enough…

Why?

• Meta has said it a hundred times already, to get to AGI you need agents and Yang le Kun is always talking about this, to the point he considers that all the fuzz about llms is distracting us from the real AI.
• OpenAI it’s been talking about agents and it is shown crearly in the essay from Sam Altman about the stages to AGI, same for Google and Demis Hassabis.
• Anthropic just released Computer use.
• Apple AI and Microsoft Copilot also have stated this is the future.
• Nvidia - Jensen Huang talks about nvidia having millions of AI workers (independent agents again) in the future.

And mainly: This is the selling point of all the huge inversions that we are watching, a 3% productivity increase does not justify these investments, Google and Microsoft building datacenters like crazy and even building SNR to be able to run them…. They are betting that we are going to have agents able to do at least part of the work unsupervised and ideally to replace part of the workforce. They want the exponencial productivity increase, first 2x, then 10x, then 100x… We’ll see if the AI companies can deliver or the Moon is farther that they thought…