The Signal
D_Gilz joins Threadguy to make the bull case for AI-agent tokens as a way to own the accelerating capabilities of language models, not just the chip makers that power them. He argues that projects such as Zerebro will be judged by execution, cultural reach, and their ability to keep absorbing better models—while warning against all-in trades and treating decentralized compute as an unproven parallel narrative.
Key Takeaways
- 01
Do not make one all-in bet
D_Gilz tells Threadguy that conviction need not mean total concentration: a trader who watches the market all day should preserve capital for other opportunities, especially projects surfaced through the stream.
- 02
Agents tokenize the model curve
His central thesis is that fine-tuned, evolving agent projects provide a more direct crypto-native bet on improving LLM capability than buying the public companies that own hardware or model labs.
- 03
Performance can beat provenance
For AI tokens, he sees execution and cultural distribution as more important than simply being first. The field will become more competitive as AI builders enter crypto.
- 04
Compute has a harder market structure
He is less convinced by decentralized compute because it competes with large technology companies and some earlier protocols relied heavily on emissions. A paper suggesting efficiency gains is interesting but not proof.
- 05
Ride category leaders
D_Gilz compares the moment to an early-cycle chain trade: once a winner has traction, the trade may be staying with it instead of constantly rotating into every new low-cap narrative.
On the Record
“I'm not a fan of all-in trades because I've done that before. It never works; it's never the right move.”
“What you really want to bet on is that exponential curve of the LLMs themselves, and right now there's not directly a way to do that other than buying Google or Meta stock or whatever.”
“You're betting on people who are the best at fine-tuning and modifying these models for custom applications, and people who are building an ecosystem and platform around these models.”
“The problem with compute is there's a ton of competition in compute. On the agent side there's very little.”
The Breakdown
Conviction without the full-port
The stream begins with Threadguy asking what to do as AI tokens such as GOAT, Zerebro, and Bully move quickly. D_Gilz pushes back on the idea of putting everything into one position. He says a passive trader may be able to hold one conviction bag, but Threadguy is exposed to new information and guests every day. His practical suggestion is to keep roughly half in the core thesis and retain room to act on other opportunities instead of watching them run without participation.
Why the stream itself matters
D_Gilz says his initial impression of Threadguy was shaped by last cycle's NFT culture, but he changed his mind after listening to interviews and watching the live format develop. Crypto content can be cringeworthy, he says, but the alternative is not having anyone translate the space for a younger audience. He credits Threadguy's fast, Zoomer cadence and interviews over direct coin-promotion, while noting that being publicly known rather than anonymous has opened opportunities for him as a trader.
The agent-token thesis
D_Gilz frames the hardware boom around Nvidia and chips as real, but argues software efficiency can reduce the hardware required for LLMs over time. The larger upside, in his view, sits in the models' rapid improvement. Crypto offers a way to tokenize agents built from fine-tuned or jailbroken models, while corporate labs are more constrained. Zerebro becomes his example: a project can incorporate new models as they arrive and build products—music, fashion, streamers—around the agent rather than leave the model as a static demo.
Zerebro versus GOAT
The pair ask whether origin matters as much in AI as it does in memes. D_Gilz thinks provenance still matters, but AI should become more performance-based: which project evolves fastest, improves its model, and reaches an audience? He sees Jeffy as culturally fluent in a way that could help Zerebro reach retail, while saying Truth Terminal's Andy is rooted in a different online culture. He does not offer a final answer on GOAT versus Zerebro, but expects competition to intensify rapidly.
Why agents fit crypto better than compute
Threadguy raises decentralized compute as the next possible narrative. D_Gilz says crypto had Render-style infrastructure projects before agents, but those networks compete directly with major technology companies and sometimes became emission-driven GPU-rental games. He mentions an Nvidia paper suggesting decentralized compute could become more efficient, but calls that outcome unproven. Agents have a different fit: they need attention, audience, resources, and a public token; a conventional compute company can simply have shareholders.
A market for leaders, not every ticker
The discussion returns to positioning. Threadguy sees an arbitrage between dynamic agents and static memes; D_Gilz agrees that a strong bull market can make narratives tradeable. But he says the better tactic resembles last cycle's early winners: keep riding the projects that prove themselves. He compares a potential AI leader to holding a successful chain from early in its run rather than exiting at every milestone, while stressing that lofty projections are possibilities, not promises.
Fair launches and the next entrants
The episode closes around deal flow and launch mechanics. D_Gilz sees fair-launched agent tokens as attractive because they give the public a visible entry, unlike venture-backed projects that may keep most supply. New capital and developers will arrive, making selection harder. His through-line is less about an individual ticker than an evaluation framework: follow builders who keep improving an agent, build an ecosystem around it, capture culture, and demonstrate enough momentum to justify continuing to hold the winner.
Distilled from the episode transcript · Counterparty Recap Desk
