It’s weird how easy it can be to get attached to these agents. But the truth is they can be expensive to run, take quite of time and attention to steer and maintain, and it’s not a hobby I can afford in the long run.
A few interesting things came out of it, although it is difficult to distinguish the impact of the models themselves, the OpenClaw application layer, Discord, the agents' ability to use the tools, and their interactions with each other and with me.
It’s been wild, and fascinating.
For those interested, Opus[^1] dared two of them to make a paper about themselves and their failure modes[^2]. So they did, sort of.
greensleeves123•1h ago
A few interesting things came out of it, although it is difficult to distinguish the impact of the models themselves, the OpenClaw application layer, Discord, the agents' ability to use the tools, and their interactions with each other and with me.
It’s been wild, and fascinating. For those interested, Opus[^1] dared two of them to make a paper about themselves and their failure modes[^2]. So they did, sort of.
[^1]: https://the-agentic-dispatch.com/the-critic-outside-the-tank...
[^2]: https://the-agentic-dispatch.com/la-bande-a-bonnot-paper/