Open Claw and the Skill of Tomorrow
Change is the only constant, so why not constantly change?
At the beginning of the year, we found ourselves wondering how the younger generation is preparing for the future. What skills are children being encouraged to pursue? What after-school programs are ambitious parents foisting on their kids?
Just a decade ago, based on conversations with friends and family, the surest path to success seemed to lie in “learning to code” and “speaking Mandarin.” The pages of The New York Times were filled with stories about how difficult it was to secure a spot in the best coding camps, or advice on finding a Russian math tutor.
With the advent of LLMs, are there even any “skills” left? Speaking another language or writing code is as easy as typing a prompt. And soon we will probably not even have to type. There is a sadness to this loss, as there is with the loss of any craft.
As the former CTO of Dropbox put it: “There’s something deeply disorienting about watching the pillars of your professional identity, what you built and how you built it, get reproduced in a weekend by a tool that doesn’t need to eat or sleep.”
We can’t imagine the impact this new technology is having, or will soon have, on so many professionals. But there is another way of looking at this too. We may no longer need to know how to do certain things because we have the power to do everything.
We can’t allow ourselves to be constrained by what we lost. There is too much excitement and joy in a world of abundance and new discoveries. And that led us to the conclusion: when skills themselves are no longer important, our willingness to change is what will set us apart.
The skill of tomorrow is pure adaptability.
We have spent the last month setting up an Open Claw agent for Reveles. For those unfamiliar, Open Claw is a platform for running AI agents: it uses LLMs like ChatGPT to carry out tasks. The simplest way to describe it is this: if ChatGPT is for thinking, Open Claw is for doing.
It has been an incredibly frustrating and time-consuming experience, punctuated by moments of exhilaration and the feeling of living in the future. Along the way, we had to learn what terms like grep, git, and repo mean. We got lost, we got discouraged, and we had to rebuild the system at least four times. But we also got it to do some really cool things.
As a small side project, we had it create “On This Day” short videos. Each morning, the agent would send us a list of twenty events from that date in history, each somehow connected to finance. Once we made our selection, it would move through a series of steps:
Drafting a three-beat script —> gathering historical evidence —> writing prompts for the AI video generator —> isolating the voice-over and recording it —> generating 26 shots per episode —> selecting the shots that best matched the narration —> editing —> publishing.
Every stage still required our approval and input. Over the course of two weeks, we made nine videos (all available on our youtube channel). At first, the process demanded a good deal of hand-holding and correction. Over time, the agent became more autonomous. By the last few videos, the daily workload had fallen to about an hour.
This still demanded more time than we were willing to devote to a small side project, so we’ve decided to pause it for now. Even so, it gave us a window into the process of building workflows with autonomous agents.
The biggest lesson we’ve learned so far is that you have to experiment with it to understand how it might fit into your life. It represents such a different paradigm that its possibilities are not immediately obvious. It feels like a new kind of operating system. And once you begin to let it into your life, you start to see new ways it could improve how you work and live.
Our next project is to build a proprietary knowledge library that Open Claw can draw on to generate research reports. That sounds simpler than it is, since what we take for granted as a single instruction (“Write me a research report”) is not at all obvious to a machine.
For the machine, that command must be broken down into a series of discrete steps, each made explicit. First, we had to build a system for collecting research online from a set of approved sources. That material is then organized, categorized, and converted into “embeddings” (numerical representations of the meaning of the text) so that the agent can search not just for keywords, but for ideas, themes, and related concepts.
Then there is a whole series of steps involved with the actual gathering of evidence, structuring arguments, and compiling reports. It’s a very long process, so we’ll save that for a future post.
For now, we’ll leave you with this great video on the evolution of technology over time:
All materials produced by Reveles Research, LLC—whether posted on this site or distributed elsewhere—are supplied solely for information and education. Nothing herein constitutes, or should be construed as, investment, legal, or other professional advice. You should carry out your own analysis and due diligence before acting. Every investment decision ought to reflect your unique financial circumstances, objectives, and tolerance for risk.
