On sharing the world with agents

This is a very interesting time in history. We are getting the tiniest glimpse of unfathomable knowledge right at our fingertips, or even as a companion we can talk to. There was even a post a few weeks ago about Brain2Qwerty, Meta’s project for converting brain signals to words. Just two hours ago, DoorDash released its CLI so agents, too, can place food orders.

All of this is happening at an explosive speed, but the cue is clear: we are going to admit agents and OpenClaws to coexist in every aspect of our lives. It’s almost a game of “can we?” rather than “should we?”

As a human living in these wild times, I give myself the right to feel weird from time to time, to take a few extra seconds on an Insta post or a song to see if it’s AI-generated or not. Do you feel me? I am capable of loving humankind, you, me, the class of 2020 in which I received my M.Sc. But I still haven’t found the time to sit and jot down my feelings about agents, which very well may be reading this post.

As an AI Engineer, I feel completely different. I know how spiky, complacent, and sycophantic the agents can be. The name of the game is “context,” and a single space can shift your response one way or another (PS, this reminded me of some research at Apple where they played with the LLM’s temperature and took the top-k candidates, and got some cool results without RL or labeled data). I personally only talk in APIs, CLIs, and agent-first lingo these days.

There’s intention, there’s execution, and then there’s this world shifting right in front of my eyes. As a guy owning both lenses, I try to build responsibly.