I want to summarize what I learn while building AI systems. I want to force myself to reflect on my lessons, and crystalize my knowledge. That is why I write this blog.
Right now my attention is fragmented. That is the first thing I have noticed. Working with several agents in parallel, or making one agent orchestrate others, makes it more diffuclt to maintain deep focus. The entire workflow is just one big distraction. You have an idea, and before distilling that thought, the best practice is to just context-dump into an agent to have it working as soon as possible so you can visualize and iterate on your idea. This workflow is a fundamental change away from thinking deeply about a problem for a while in order to solve it, as the most efficient way to judge a proposed solution is to see it in practice from an agent.
Attention gets fragmented by the workflow, but learnings are also in a fluid state. Everything is based upon rapid iteration, back and forth, and intuitive decisions. Very rarely do you learn a concrete thing, it is most often just intuition being built for what the agents can and can’t achieve, and what tone of prompt best would achieve this. This is why i try to articulate them here; in order to crystalize the intuitions.
I hope this fragmentation is just a temporary symptom. Right now the agent-swarms are just not self contained enough that you can fire and forget. If you write a good spec, you still won’t get it tailored exactly to your needs. Improvements to the systems may help the human disconnect from technology and move back into its head for deep thinking. However, this workflow is unlikely. You would probably still want an LLM to help you iterate the spec. They are just that efficient at producing stuff. Maybe one has to conciously remove them from ones workflow to preserve sanity.
There is absolutely leverage to be had by being the greatest at AI tools, and working with them intimately. But there is also a huge cost; focus. The great CEOs lead fragmented work-lifes, like all of us soon will. I suspect the winners will be the ones that manage to preserve their deep focus.