Skip to main content

AI: The Gift that Never Takes

AI Dispatch #44: Best of December

Castle Labs3 min read
AI: The Gift that Never Takes

For one of the last weekly Dispatches of the year, we return to our old format.

While we take some time to reflect and close the year, morale seems to be fairly low in the markets, as reflected by the fact that AI as a whole is one of the losers in mindshare, as mentioned by @KaitoAI.

However, we should take this occasion to reflect on how far we have come.

Especially for AI, 2025 was the year of mainstream establishment, and having practical implementations of agentic communication and commerce in production.

One of the main proponents of agentic activities has been @virtuals_io with its ACP.

Just a few days ago, they hit over $370m processed from AI agentic activities:

Article figure

This was also the week of @solana breakpoint, where @jerallaire from @circle talked about their vision for AI in the next 3-5 years:

Article figure

Whenever we talk about AI agents, one of the first questions that people ask is, can they make me money? Here’s an experiment on the matter by@diego_defai across @ZyfAI_ @SurfAI @almanak @gizatechxyz and more.

Article figure

Surely AI agents continue to be appealing to investors:

@SurfAI recently announced their $15m funding round:

Article figure

One of the aspects of AI that will be critically important on its development is regulation. Just a couple of days ago, Trump signed an executive order to ensure a national policy for AI, prohibiting states from regulating AI at the state level.

Article figure

Here’s a deeper dive into the reasoning behind:

Article figure

The end of the year is a great time for content coming out (we will have something for you as well, coming soon).

@a16crypto's big ideas include a few on agents and AI:

Article figure

How do we make AI verifiable? And what does it mean?

@eigencloud shares their vision through a case study with @StoryProtocol:

Article figure

For a deep dive into this, you should listen to @therollupco interview with @sreeramkannan and @devrelius:

Article figure

In other news, @SentientAGI just realised their “Semantic Embedding and Reasoning Agent” (SERA-Crypto), which appears to have outperformed most of the current LLM models such as GPT-5, Grok 4, Gemini 2.5 Pro and Perplexity.

Article figure

After a period of radio silence @Punk9277 has released a longer post about the future of InfoFi and @KaitoAI, worth a read:

Article figure

Totally a coincidence that this post comes right after, with some light-hearted criticism on the state of AI and what we achieved so far: expectations vs reality, I guess.

Article figure

JKJK... Slowly but surely, AI will penetrate more and more of our flows. Here’s @DeFiMinty with some useful tools for research:

Article figure

To be completely honest, at Castle, we really use AI marginally, and always as a tool rather than a means of its own.

We follow up with some interesting insights by @JaredKubin on LLMs performance IRL vs in controlled environments:

Article figure

Now onto a different type of use of AI.

How does @altitude leverage AI? We can expect more “neobanks” and similar to introduce these implementations.

Article figure

Also, some AI updates from the @HyperliquidX ecosystem, with a dive into the technical challenges to implement “mirror trading” on Hyperliquid.

Article figure

We end up on a funny note with @saylor being the only one who consistently managed to turn AI slop into its own branding and still does it. Well gg! Our timeline is ruined but at least we can laugh about it.

Article figure

Last but not least, we end with a post from @0xJeff bringing some much-needed perspective on the evolution of AI within crypto + some predictions for the future.

Article figure


Some Little Web2 AI Bonus News:

Article figure

  • Smart enough? Pretty good occasion to join the @AnthropicAI fellows program.

Article figure

  • Another pretty sick example of AI for video, with the @invideoOfficial AI engine able to replace actors and scenes while keeping their expressivity.

Article figure

  • Here’s an interesting image on the current flows among AI companies:

Article figure

  • The more data, the more problems - never thought we’d be sharing a post by Mario Nawfal, but here we are. Pretty crazy how Netflix’s immersive experience turned into a data harvesting operation.

Article figure

That’s it for this week, we wish you a peaceful Monday and a great rest of the week.

Originally published in the Castle Labs newsletter. Subscribe at research.castlelabs.io.