Author: Haseeb, Dragonfly Partner; Compiler: 0xjs@黄金财经
I will either look like a prophet or an idiot, but one thing is for sure: I will annoy a lot of people with bags.
I will divide my predictions for 2025 into six parts: L1/L2, token issuance, stablecoins, regulation, "AI agents" and Crypto x AI.
1. L1/L2
The distinction between L1 and L2 is disappearing.Users can no longer perceive the difference between L1 and L2 (did they ever perceive it?). The blockchain field (L1 and L2 combined) is already too crowded and needs a major reshuffle. Consolidation will no longer be about technological advantages - it will be about owning a unique niche market and building stickiness through GTM.
Despite the strength of SVM and Move, EVM market share will actually grow in 2025. This growth will be driven by Base, Monad, Berachain. This is no longer because of compatibility, but because EVM/Solidity has more training data, and LLM will write most application code in 2025. Having deep, battle-tested crypto contract libraries will also be a watershed, because LLM are not good at writing low-level code. In the era of LLM development, DevEx and footguns are less important than training data and reliable libraries.
Solana will force more blockchains to optimize for low latency. We will move from a TPS war to a latency war - infrastructure like doublezero and ultra-low latency L2s like MegaETH will drive user expectations for web2 responsiveness. Expect more people to embrace optimistic UI, pre-confirmation, intent, email onboarding, in-browser wallets, and progressive security. The emergence of swap privy. Hyperliquid has proven that when specialized chains focus on specific applications and prioritize user experience and easy bridging, they can work. More projects will follow this model. The old dream of one chain to rule them all is dead. Token issuance The era of massive airdrops through point programs is over. We are moving towards a two-track world. Track 1: If a project has a clear North Star metric, such as an exchange or lending protocol, they will distribute tokens purely based on points. They won’t care if they are being farmed or gamed — they are essentially distributing tokens as kickbacks/discounts on the core KPIs of the protocol, and airdroppers are your actual users anyway.
Track 2: Projects without a clear North Star metric (like L1 and L2) will turn to crowdfunding. They may do small airdrops to reward social contributions, but most tokens will be distributed through crowdfunding. Airdrops for vanity metrics are dead. Those that are not going to real users are going to industrialized airdroppers.
Memecoins will continue to lose market share to “AI agent” coins. I see this as a shift from financial nihilism to financial over-optimism. (Yes, I coined that term.)
3. Stablecoins
The use of stablecoins will surge, especially among small and medium-sized enterprises. Not just trading and speculation — real businesses will start using on-chain dollars for instant settlement.
Banks are taking notice: expect to see announcements of bank-issued stablecoins by the end of 2025. They don’t want to be left behind. But especially with Lutnick as Commerce Secretary, Tether will retain its No. 1 spot.
Expect Ethena to devour more capital, especially as Treasury yields continue to fall over the coming year. When the opportunity cost of capital falls, it makes the underlying trading yield more attractive
4. Regulation
The U.S. passes stablecoin legislation while broader market infrastructure reform (FIT21) is delayed. Stablecoin adoption accelerates while Wall Street adoption, asset tokenization, and other TradFi integrations lag behind.
Under Trump, Fortune 100 companies will be more willing to offer cryptocurrencies to consumers, with tech companies and startups showing a higher appetite for risk. Trump’s inauguration will create a palpable regulatory celebration until clear rules and enforcement priorities are established. In the interim, expect to aggressively expand the integration of cryptocurrencies with Web2 platforms.
5. AI Agents (This is the longest section because my ideas here may be controversial — read to the end!)
The “AI Agent” craze may last until 2025. But it will eventually die.This is not the long-term disruption of AI that needs to be vigilant, but it will be a concern for CT because it is the most social.
Current AI agents aren’t really intelligent agents. These are chatbots with memecoin attached; they have little to no intelligence other than posting on Twitter. Current “AI agents” are also mostly “Wizard of Oz” agents — there are humans behind the scenes making sure the AI doesn’t go off the rails. This isn’t going to change anytime soon, because current agents are so bad (even Fortune 100 companies aren’t using agents in production yet). Current agents can easily be manipulated into saying crazy things that damage their brand, or can be jailbroken to have all of their resources stolen. See Freysa for what truly autonomous AI looks like — if your favorite AI isn’t jailbroken, it’s because it’s Wizard of Oz AI.
That being said, I think this trend will accelerate. Chatbots can indeed replace a lot of KOLs because chatbots never sleep, they are always delivering information, and they are not as greedy as human KOLs. Also, most KOLs are not very creative. Even today, real-time information aggregation/amplification can be easily replaced by algorithms (see @aixbt_agent)
Now these chatbots are very attractive to us because they are very novel. It's like seeing an elephant painting a picture. The first time you see it, you don't care if the painting is good or not - it looks spectacular. But the 1000th time, the novelty wears off. I believe this will start to happen as these chatbots reach a steady state.
You can see this with aixbt today - it's already very good at aggregating data about different projects. By next year and the next generation of agents, maybe aixbt will be a little less hallucinatory, a little deeper, and have a smarter view. But how much will you notice? For most people, it will probably feel the same.
I think this sense of novelty and market desire will continue until 2025. It will take a while for cryptocurrencies to get tired of the shiny stuff. But by 2026, I think there will be a sudden reversal. Chatbots will become so ubiquitous that people will lose interest in them. Sentiments will reverse. Seeing stories of their favorite human influencers losing their livelihoods will inspire a class consciousness. Users will start discriminating against human influencers, even if their content is less consistent.
To counter this pro-human bias, chatbots will start hiding that they are AI, trying to pass themselves off as humans to attract more of the attention market. Instead of making money through memecoin like they do today, future chatbots will make money like human influencers do through sponsorships, affiliate links, and tokens they own. Influencers will be accused of being chatbots regularly, and you’ll see scandals revealing AIs. It’s all going to get weird.
But there’s a darker side. Keep in mind that LLMs are currently good wordcels, but not good enough for other things. What is the best way to make money as a wordcel in crypto? First is to become an influencer, sure, but a close second is to become a scammer. You will start to see a surge in autonomous scam bots. These will explode in popularity, comparable to ransomware and cryptojacking after 2017. Expect this to become a real social problem.
However, while chatbots will likely continue to be a focus in 2025, the long-term disruption of AI will not be on the social level.
No, it won’t be in the trading space either. AI will not give everyone their own "trading agent" or micro-hedge fund. Yes, AI will scale everyone, but they will scale people in proportion to capital, data, and infrastructure. So you should expect AI to supercharge existing trading firms that have capital scale and data scale. In other words, trading firms will get better at making money. It will also break down the hierarchy among trading firms (most trading firms will become pretty good because everyone has access to a quant with an IQ of 150 in the cloud).
Over time, AI will make markets extremely efficient, even in smaller niche markets, which will leave little advantage for the average trader, even if they have a homemade assistant AI. The value of original research will drop dramatically. Still, the increase in competition and liquidity should be a boon to those of us who inject noise into the market. (That means liquidity for Polymarket, too!)
So if the big news isn't chatbots and it isn't trading bots, then what is it? Here’s my core thesis, and for some reason almost no one is talking about it: The truly impactful AI agents will be software engineering agents.
Why is this a big deal? Ask yourself: What is the primary input for our industry? What expensive input is holding back more applications, more wallets, better infrastructure, better everything? The answer is software. If AI agents cause software prices to collapse, that will change everything.
In the post-AI world, instead of raising millions for a seed round, you’ll be able to launch an application using $10k of AI cloud computing. Self-funded projects like Hyperliquid and Jupiter will go from being the exception to the norm. The number of on-chain applications and experiments will absolutely explode. For an industry driven by software, this deflationary shock will lead to an on-chain renaissance.
The implications for security are profound. AI-driven static analysis and monitoring will be ubiquitous, making security more accessible to everyone. These AIs will be fine-tuned on EVM/Solidity or Rust codebases and trained on a vast database of security audits and attack vectors. They will be trained with RL in simulated adversarial blockchain environments. I am increasingly convinced that AI tools will ultimately favor defenders over attackers when it comes to security. You will have AIs that are constantly red-teaming contracts, while other AIs will harden them, formally verify their properties, and hone their skills in incident response and remediation.
In the meantime, of course, AI-style memecoins can be traded. But real agents will have a much greater impact than tweeting and promoting their own tokens.
6. Real Crypto x AI
Above I detailed the impact of AI on cryptocurrency (which is the main direction of impact), but cryptocurrency will also have an impact on AI.
True autonomous agents will pay each other using cryptocurrency. This will be especially true once there is relaxed stablecoin regulation — you’ll even see large companies running AI agents using stablecoins for inter-agent payments as they’re easier to boot up than bank accounts.
We’ll also see more and larger experiments with decentralized training and inference. A new generation of promising projects like Exo Labs, Nous Research, and Prime Intellect will pave the way for real alternatives to centralized training and corporate self-operation models. NEAR Protocol is also working full-throttle to create a full-stack, trusted neutral, and permissionless AI stack.
Another place where Crypto and AI intersect is in user experience. Post-AI wallets will revolutionize — AI wallets should be able to handle bridging, optimize transaction routing, minimize fees for you, mask interoperability issues or front-end errors, and help you steer clear of obvious scams or rugpull. You won’t have to switch between multiple different wallets, change RPCs, or rebalance your stablecoins. The AI will handle all of that for you. It may take until 2026 before this becomes reliable enough to change the user experience of crypto. But when it does, what will this do to blockchain network effects? What happens when users no longer care — or even experience — which chain an application exists on?
This space is still young, but I expect we’ll see it take off soon. In the long term (like mid-2026), I expect the majority of the “AI x Crypto” market cap to be located here. -
That’s all my predictions. Hopefully this time next year you won’t have to work!
Alex
Weiliang
Miyuki
Brian
Joy