Written by: Haseeb Qureshi, Partner at Dragonfly, Translated by: Yangz, Techub News After reading these predictions, you might think I'm either a prophet or an idiot, but one thing is certain, these predictions will annoy many people. My predictions are divided into six parts: L1/L2, token issuance, stablecoins, regulation, AI agents, and the intersection of cryptocurrency and AI. 1. L1/L2 The distinction between L1 and L2 is disappearing, and users no longer perceive the differences between L1 and L2 (did they ever?). Looking across the current state of the blockchain industry, there are countless L1 and L2, and a reshuffle is imminent. Integration will no longer be about technical advantages, but about having a unique niche market and building stickiness through GTM (Go To Market). Despite the strong momentum of SVM and Move, the market share of EVM will still grow by 2025. This growth will be driven by Base, Monad, and Berachain. This growth is no longer due to compatibility, but because EVM/Solidity has more training data, and by 2025, most application code will be written by LLMs (Large Language Models). Furthermore, whether one has a deep library of battle-tested crypto contracts will also become a watershed, as LLMs are not good at writing low-level code. During the era of LLM development, the importance of DevEx and footgun (usually refers to programming that is prone to errors or security risks) will be lower than that of data training and reliable codebases. Solana will force more blockchains to optimize for latency performance. We will shift from a TPS war to a latency war. Infrastructure like Doublezero and super low-latency L2s such as MegaETH will drive user expectations for Web2 response times. Friendly user interfaces, pre-confirmations, intent, email login methods, in-browser wallets, and progressive security will be more favored. In this regard, Web3 authentication and private key management tool Privy sets a good example. HyperliquidX has already proven that dedicated chains focusing on specific applications and prioritizing user experience and seamless cross-chain functionality are viable. In the future, more projects will adopt this model. The old dream of a single chain ruling them all has collapsed. 2. Token Issuance The era of heavily airdropping tokens through points programs is over. We will enter a 'dual-track world'.
Track 1: Projects with clear metrics, such as exchanges or lending protocols, will distribute tokens purely based on points. They will not care whether the tokens are 'farmed', but will distribute tokens as rebates/discounts based on the core KPI of the protocol, with farmers being the actual users.
Track 2: Projects without clear metrics (such as L1 and L2) will turn to crowdfunding sales. They may conduct smaller-scale airdrops to reward social contributions, but most tokens will be distributed through crowdfunding. The form of airdropping tokens for vanity metrics is dead. There are no real users under these metrics, only industrialized farmers.
Moreover, the market share of Memecoins will continue to be replaced by 'AI agent' coins. I believe this is a shift from financial nihilism to financial excess optimism. (Yes, I coined this term myself.)
3. Stablecoins
The use of stablecoins will experience explosive growth, especially among small and medium-sized enterprises. Not only for trading and speculation, but real businesses will also start using on-chain dollars for instant settlement.
Additionally, banks are also taking note of this, and to avoid falling behind, it is expected that by the end of 2025, some banks will announce the issuance of their own stablecoins. However, Tether will still be the leader in the stablecoin space, especially after Lutnick takes over as Secretary of Commerce. (Translator's note: The Trump administration nominated Howard Lutnick, CEO of Wall Street bond trading giant Cantor Fitzgerald, to be the next Secretary of Commerce. Cantor Fitzgerald has been custodians of Tether's assets since the end of 2021.)
Furthermore, as government bond yields continue to decline over the next year, Ethena is expected to absorb more capital. When the opportunity cost of capital decreases, the base trading yield becomes more attractive.
4. Regulation
I expect that the United States will legislate stablecoins by 2025, while broader market infrastructure reform (FIT21) will be postponed. I believe the adoption of stablecoins will accelerate, while Wall Street's adoption, asset tokenization, and other TradFi integrations will lag behind.
Under the Trump administration, Fortune 100 companies will be more willing to offer cryptocurrency to consumers, and tech companies and startups will exhibit a higher risk appetite. Before clear rules and enforcement priorities are established, Trump's inauguration will create a 'regulatory carnival'. During this window, cryptocurrency will be actively integrated into Web2 platforms.
5. AI agents
This section is the longest, as my ideas may be controversial, so please read to the end.
The hype around AI agents may last until 2025, but it will ultimately fade away. This is not a long-term disruption that the AI field needs to be wary of, but it will become a focal point for CT, as it is the most social.
The current AI agents are not true agents. They are actually just chatbots that issue Memecoins. Aside from posting on Twitter, they have almost no agency functionality. Moreover, the current AI agents are mostly 'Wizard of Oz' agents, with humans ensuring that AI does not go out of control behind the scenes. This situation will not change in the short term, as the current agents are not very 'intelligent' (even Fortune 100 companies have not yet used agents in prototypes). They can be easily manipulated to say crazy things that damage their brand or be hacked to steal all their resources. You can look at Freysa ai to understand what true autonomous AI looks like. If your favorite AI has not been hacked, it is because it is a 'Wizard of Oz' style AI.
Translator's note: In the field of human-computer interaction, 'Wizard of Oz' is a research experiment where subjects interact with a computer system that they believe is autonomous, but is actually being operated or partially operated by an unseen person.
Nevertheless, I believe this trend will accelerate. Chatbots can indeed replace many influential people because they do not need to rest, will constantly send messages, and they are not as greedy as human KOLs. Additionally, most influential individuals are not known for their originality. Even today, real-time information aggregation/amplification can easily be replaced by algorithms (such as the aixbt agent).
Right now, we are fascinated by these chatbots because they are so novel. Just like seeing an elephant painting, at first glance, we don’t care whether the artwork is exquisite, but after the 1000th look, the novelty wears off. I believe that as these chatbots stabilize, this situation will begin to manifest.
We can see this from aixbt, which has already excelled at aggregating data from different projects. By next year, when the next generation of agents emerges, maybe aixbt will reduce misinformation generation (also known as 'hallucination'), analyze more deeply, and make smarter trade-offs. But how much can we notice? Most people's feelings might not differ from now.
I believe this novelty and market desire will last throughout 2025, after all, the cryptocurrency industry usually takes time to tire of these shiny things. By 2026, I think the situation will suddenly reverse. Chatbots will become so ubiquitous that people will become averse to them. People's sentiments will shift. When observing stories of their favorite human KOLs losing their livelihoods, a class consciousness will ignite in people's hearts. Users will dialectically begin to lean towards human KOLs, even if the content produced by these KOLs is inconsistent.
To counter this tendency towards human bias, chatbots will begin to hide the fact that they are AI, attempting to impersonate humans to garner more market attention. Future chatbots will no longer profit like they do now through Memecoins, but will earn like human KOLs, through sponsorships, affiliate links, and holding their own tokens. By then, human KOLs will frequently be accused of being chatbots, and we will see many news articles revealing the 'AI mask'. All of this will become quite strange.
However, behind this, there is a darker side. Remember, LLMs are currently great at text processing, but they are not mature in other aspects. In the cryptocurrency field, a role that excels in textual expression will find that the best way to monetize is first to become an influential person, and secondly to become a scammer. We will see a massive surge of autonomous scam bots, which will rival the ransomware and crypto hijacking of the post-2017 era. In my view, this will become a real societal issue.
However, that said, while chatbots may still be the focus of attention in 2025, the long-term disruption brought by AI will not manifest at the social level. Of course, it will not be in the trading field either.
AI will not give everyone their own 'trading agents' or micro hedge funds. Yes, AI can enhance everyone’s capabilities, but they will empower based on people's capital and other strengths. Therefore, we will see AI injecting more momentum into existing trading firms that have capital scale and data scale. In other words, trading firms will become better at making profits. Additionally, AI will break down the hierarchy between trading firms (most trading firms will become quite proficient since they all have access to cloud-based data analysis experts with an IQ of 150).
Over time, AI will make the market extraordinarily efficient (even in smaller niche markets), and the advantage of ordinary traders will disappear (even if they have their own AI assistants), while the value of original research will drop sharply. Nevertheless, the intensified competition and increased liquidity should be a boon for those of us who inject noise into the market.
So, if the core of AI agents is not chatbots or trading bots, what will it be? In my view, truly influential AI agents will be software engineering agents. This is my core argument, but for some reason, almost no one talks about it.
As for why software engineering agents will have a significant impact, one might ask themselves, what is the main input in our industry? What hinders more applications, more wallets, better infrastructure, better everything? The answer is software. If AI agents lead to a drastic drop in the cost of building software, everything will change.
In the post-AI era, project teams will no longer need to raise millions of dollars for seed funding, but only spend $10,000 on AI cloud computing to launch an application. Self-funded projects like Hyperliquid and Jupiter will become the norm rather than the exception. The number of applications and experiments on-chain will explode. For a software-driven industry, this shock of cost deflation will trigger a renaissance on-chain.
Moreover, the impact on security will also be profound. AI-driven static analysis and monitoring will become ubiquitous, making it more convenient for everyone to access security information. These AIs will fine-tune on EVM/Solidity or Rust codebases and will be trained on vast security auditing and attack vector databases. They will undergo reinforcement learning (RL) in simulated adversarial blockchain environments. I am increasingly convinced that, in terms of security, AI tools will ultimately benefit defenders rather than attackers. We will see AI constantly 'red-teaming' contracts, while other AIs will harden contracts, formally verify their properties, and hone their incident response and remediation skills.
During this time, it is normal for trading AI-flavored Memecoins to emerge, but true agents will be more influential than 'pumping' their own tokens on Twitter.
6. True integration of cryptocurrency and AI
In the above text, I detailed the impact of AI on cryptocurrency (main impact direction), but cryptocurrency will also impact AI.
True autonomous agents will use cryptocurrencies for mutual transactions. Once there are relaxed stablecoin regulations, this will become particularly apparent. We will even start to see large companies operating AI agents using stablecoins for inter-agent payments, as stablecoins are easier to establish than bank accounts.
Additionally, we will see more larger-scale decentralized training and reasoning experiments. Organizations like EXO Labs, Nous Research, and Prime Intellect will pave the way for truly replacing centralized training and company-owned models. And the NEAR Protocol is trying to create a fully trusted, neutral, permissionless AI stack.
Another intersection of cryptocurrency and AI is user experience. The wallets of the post-AI era will be radically different. An AI-driven wallet should be able to handle cross-chain transactions, optimize transaction routes, reduce fees for users, resolve interoperability issues or front-end errors, and guide users away from obvious frauds or scams. Users will no longer have to switch back and forth between multiple different wallets, nor will they need to change RPCs or rebalance stablecoins; AI will handle all these issues for users. However, this may take until 2026 to become reliable enough to change the cryptocurrency user experience. When all this arrives, what will the network effects of blockchain look like? What happens when users no longer care, or even perceive which chain an application is on?
This field is still in its early stages, but I hope we will soon see its takeoff. In the long run (say by mid-2026), I expect most of the market value of 'AI x cryptocurrency' to concentrate in this field.
These are all my predictions. I promised I would write these down before reaching 100,000 fans, so I’m a bit late, but I still managed to complete it within the New Year! Disclosure: These are all my personal views and do not represent Dragonfly; many companies mentioned in this article are invested in by Dragonfly; no financial advice is provided, DYOR. I could be AI, or I might not be.
