Good writeup Kyle. I actually disagree with you on the innovation piece. I think there are so many vectors projects can take in terms of pushing forward new agents in interesting novel ways. And as you rightfully pointed out, those will be pushing the envelop for the "new meta"
Projects I have yet to but want to see:
- Agents that can speak to you using an e2e voice agent and llm on the backend; this can be through a phone call
- SOTA video agent - with all the new video models (Google's Veo2 and
@zhou_xian_'s Genesis), there should be insane agents that ship the frontier
- Agents that ACTUALLY make money - classical ML legitimately predicts the future by using data today, this is the premise of quant trading using algos - combine that with a shitposter KAIL (key AI leader), where users can deposit money, and the agent can tweet their trades / thought process. Now, that's interesting
- New model architectures. There is this one project that I follow closely that uses a completely in-house framework for visual language models that continuously iterates and improves images through internal reasoning
- Cross-chain agents - this is what
@0xzerebro is building. You saw what they have been doing with
@base and
@LayerZero_Core. It's incredibly exciting
- Agents that can start autonomous conversations with you in DMs over X
- Agents that measure your X influence, and analyze your profile, and give it some sort of quantitative score (cough
@SHL0MS @god @s8n)
- More fine-tuned agents with sharper, deeper personalities. Can thank
@NousResearch for this one
- Agents that utilize many models but autonomously pick the best one based on the previous post, or a reply - essentially routing
I could go on, but my point is that the surface has only been scratched. The tech is still so so so early, and if you're building any of the above, please reach out. Would be happy to chat over the holidays