Artificial intelligence in cryptotrading: Between hype and reality. What awaits the market in 2025-2035?

What’s going on? Navigation in the fog of war

Let’s get straight to the point. I won’t bother you with the formalities. The purpose of this analysis is to give an honest look at what the world of AI-cryptotrading is now. And it is a seething cauldron of breakthrough technologies, colossal risks and, frankly, wild marketing. Here, innovation goes hand in hand with technological anarchy, and regulators are already looming around the corner. Go.

My main thought

AI in cryptotrading is not the future, it is already the present. And it is driven mainly by Generation Z, for whom it is as natural to entrust money to a bot as it is for us to order pizza through an application. But this rapid growth has created a wild, chaotic market. The technological “arms race” is in full swing, but I have long come to the conclusion that long-term success here does not depend on the coolest algorithm. It depends on whether we can solve three fundamental problems: safety, reliability, and regulatory chaos.

If you don’t have much time: the main thing in a nutshell

• A market that no one can calculate: Ask five analysts what the size of the AI trading market is, and you will get five answers that vary by an order of magnitude – from $1.46 billion to $40.8 billion. This is not an error, it is a symptom of wild immaturity and lack of uniform rules of the game.
• Epoch Z: Young traders are the main driver of growth. For them, bots are not just a way to make money, but something like an “emotional crutch” that helps them not to go crazy on the roller coaster of the crypto market.
• Technologies under the hood: Good old hybrids like CNN-LSTM are still in service. Transformers and Reinforcement Learning (RL) are at the forefront of the attack, but they are moody and often “stick” to the market noise. It seems to me that the future belongs to systems that can predict, make decisions, and calculate the moves of competitors.
• Where they play: Commercial platforms (Bitsgap, 3Commas) beckon with convenience, but, as practice has shown, they can be leaky in terms of security. Open-source projects (Hummingbot) give you full control, but require you to be a bit of a programmer.
• Alpha is a melting ice cube: Any trading advantage (“alpha”) evaporates over time. This is the law of the market. Success is not about finding one “grail”, but building a conveyor belt for the production and replacement of these “melting cubes”.
• The End of the Wild West: The regulators have woken up. The SEC and CFTC are starting to crack the whip, punishing empty marketing about “AI” and making it clear that the old rules also work for new toys.
• Big game: The idea of “sovereign AI”, when each country wants to control its technology, comes into direct conflict with the idea of decentralized cryptocurrencies. This creates an interesting puzzle.: which will win — decentralization, productivity, or national interests?

The paradox of market valuation: How to measure a ghost?

The first thing that catches your eye when you start digging is the utter mess in the estimates of the market size. The figures published by reputable research firms look like they describe different planets. And this is not just a fun fact, it is the main symptom of the fact that we are dealing with a nascent and absolutely chaotic sector.

• WiseGuyReports talk about $1.46 billion in 2024, with an increase to $25 billion by 2035.[1] At the same time, Research and Markets are painting a completely different universe: $40.8 billion in 2024 and almost a trillion ($985.2 billion) by 2034.[2] [3] [4] Market.us they are trying to find a middle ground, including mining hardware in the calculations, and receive $3.7 billion in 2024.[5]
• And Lucintel have apparently decided to be the most cautious and forecast a modest $0.17 billion by 2030.[6]

Where does this “fog of data” come from? It’s simple: everyone thinks in their own way. Someone includes video cards and servers in the evaluation (which, according to the data Market.us , account for almost half of the market [5]), someone only looks at the software, someone-some are for retail, and some are for institutions. For any investor, this means one thing: blindly believing one report is suicide. But for those who know how to think with their own heads, this is an opportunity.

Research Firm	Market Definition	Assessment (2024)	Forecast (Year / Size)	CAGR
WiseGuyReports	Bots for AI-based Crypto Trading	$1.46 B	2035 — $25 B	29.5%
Research and Markets	AI Crypto Trading Bot Market	$40.8 B	2034 — $985.2 B	37.2%
Market.us	AI Crypto Market (incl. Hardware)	$3.7 B	2034 — $46.9 B	28.9%
Lucintel	Bots for AI-based Crypto Trading	–	2030 — $0.17 B	32.4%

Table 1: Comparative analysis of forecasts for the size of the AI-cryptotrading market But here’s what’s interesting. This leapfrog with numbers is not just noise. It shows where different players see future money. Those who include hardware in calculations [5], they are betting on an infrastructure war, on “picks and shovels”, where NVIDIA and AMD rule. Those who focus on software [1], [2], believe in “the gold diggers.” So the size of the market is not a given that needs to be found, but a story that needs to be told. And whoever tells it more convincingly will attract more money and talents.

New players, new rules: Psychology of Generation Z

So, we found out that the market is chaos. Now let’s see who is voluntarily participating in this chaos and why. It’s all about generational change, and it changes the rules of the game much more than any new technology.

Generation Z: Digital Aborigines in the Wild West

Forget about the wolves of Wall Street in expensive suits. The main consumer of AI bots today is a young trader, for whom a smartphone is an extension of his hand, and automation is second nature.

• According to MEXC Research, which analyzed almost 800,000 users, 67% of Generation Z traders (18-27 years old) we turned on the AI trading bot at least once.[7] [8] Think about it: two out of three.
• These guys account for 60% of all bot activations. They use them on average 11.4 days a month — twice as often as traders over 30. href=”https://www.globenewswire.com/news-release/2025/07/24/3120851/0/en/MEXC-Research-Every-Second-Gen-Z-Trader-Now-Relies-on-AI-for-Crypto-Trading-Decisions.html ” target=”_blank” rel=”noopener”>[7] [8]

This is not just a trend. This is a fundamental shift. Young traders are “AI-native”, they think in terms of automation. While millennials are still drawing lines on graphs, Generation Z is simply delegating the routine to the machine.[7] [8]

The psychology of automation: Bot as a sedative

But here’s the most interesting thing: it’s not just about laziness or technological advancement. I am sure that for Generation Z, bots are primarily a way to keep your sanity in a market that is trying to drive you crazy 24/7.

• Informed delegation: They don’t just blindly trust the bot. 73% of users turn it on during wild volatility and turn it off when the market is calm. It’s not a belief, it’s a tactic.[7]
• Emotional anchor: And here’s the real kicker. The use of bots reduces the number of panic sales by 47% during market crashes.[7] [8] A bot is a cold, emotionless machine that holds your hand when you want to press the red button.
• Self-protection: Bot users are 1.9 times less likely to make impulsive trades and 2.4 times more likely to place stop losses and take profits.[7] [8] In fact, they hire AI to impose discipline on them, which they themselves lack.

For these guys, AI is not just a tool, it’s a team partner. Research Resume.org It showed that more than half of young workers consider ChatGPT to be their colleague [7]. The same logic applies here. This changes everything for platform developers. Switching the bot on/off is not just a technical action. It is the act of “releasing” your AI partner onto the field when the game becomes too difficult. This is delegating not only work, but also stress. Trust here is not a blind faith that a bot will bring millions. This is the trust that he will coolly execute the plan while you go to drink coffee, so as not to do anything stupid. Platforms that understand this psychology of collaboration will benefit. Their interface should be similar not to a broker’s terminal, but to a shared workspace with smart assistants.

The arms race: What do they have under the hood?

All this is good, but what does it all work on? Let’s take a look into the engine room without the marketing husk and see exactly what technologies are trying to tame the crypto market today. And why it’s so damn difficult.

The problem at the entrance: Why financial markets are hell for AI

Predicting prices is a thankless task. Especially in the crypt. Financial data is not like cats in pictures to recognize, they have several innate properties that break standard AI models.

• Key concepts: Imagine an ever-changing landscape (unsteadiness) where a useful signal sinks into an ocean of random noise (low signal-to-noise ratio), and all this happens 24/7 with wild price spikes (high volatility).[9] [10] [11] [12] [13] [14] [15] [16] [17] Simply put, the rules of the game are constantly changing, and the model that worked yesterday can drain your deposit today.
• Unique crypt data: On the other hand, we have data that is not available in traditional markets: on-chain metrics, data from order glasses, sentiment from social networks. This provides more clues, but also makes the task an order of magnitude more difficult.[18] [19]

Workhorses: The Old Guard of neural Networks

Despite all the hype around new architectures, the basis of many systems today are time—tested hybrid models.

• Hybrids (CNN-LSTM): It’s like a combination of two specialists. CNN (convolutional neural network) is a “graphic technician”, she finds familiar patterns on the chart perfectly, like an experienced trader. A LSTM (long-term short-term memory network) — This is a “historian”, she understands the context and remembers what happened before., capturing long-term trends.[11] [20] [21] [22] They work surprisingly well together.
• Performance: Time after time, research has shown that this pair is bypassing both CNN and LSTM individually. They are reliable enough to be the basis for many strategies.[20] [23]

At the forefront of the attack: Transformers and RL

And here the most interesting and risky thing begins. These are technologies that promise revolution, but so far more often bring headaches.

• Transformers: These guys come from the world of language processing (models like GPT work for them). Their superpower is the attention mechanism, which allows them to see connections between very distant events in the data, which is more difficult for LSTM.[9] [24] But there is a huge problem.: They are incredibly “voracious” about data, and in a noisy financial market they easily begin to see patterns where there are none. This is called retraining, and it is the main enemy of any quantum trader.[9] [25] [26]
• Reinforcement Learning (RL): This is a different approach altogether. Instead of predicting the price, the RL agent learns to trade like a human being through trial and error. He makes trades and receives a “reward” for profit or a “penalty” for loss, gradually developing an optimal strategy.[27] [28] [29]
  • DQN: It is good for simple tasks: buy, sell, hold. But when you need to decide how much to buy, he gives up.[28] [29]
  • PPO: is a more advanced and stable method. He can work with continuous actions, for example, decide what percentage of the portfolio to invest in a deal. Therefore, it is more suitable for serious portfolio management.[27] [28]

Architecture of the Model	Main Mechanism	Main Advantage	Key Weakness / Problem	Ideal Use Case in Cryptotrading
CNN-LSTM	Hybrid of convolutional and recurrent layers	Combines local pattern extraction (CNN) with time dependence modeling (LSTM)	Less effective for very long-range dependencies compared to Transformers	Medium-term trend forecasting combining chart patterns and temporal dynamics
Transformer	Self-Attention mechanism	Captures long-range dependencies and allows high parallelization	Requires large datasets; prone to overfitting on noisy market data	Analyzing macroeconomic news or events requiring long-range context
DQN (RL)	Q-function learning (action-value estimation)	Effective for discrete actions (buy/sell/hold)	Limited flexibility — struggles with continuous actions (e.g., position sizing)	Simple automated bots reacting to indicator signals
PPO (RL)	Policy optimization with step limitation	Stable learning, works well in continuous action spaces	Slower convergence compared to value-based methods	Dynamic portfolio and risk management with continuous position control

Table 2: Comparative overview of AI model architectures for financial forecasting

What’s next? Looking beyond the horizon

All this is already today. And what awaits us tomorrow? Researchers are digging in several directions, and some of them look very promising.

• Causal AI (Causal ML): This is an attempt to teach AI to distinguish a simple correlation from a real causal relationship. For example, to understand that it is not the sunrise that makes the rooster crow. This is the holy Grail for trading: to find the real market drivers., not random coincidences.[30] [31] [32] [33] [34] [35]
• Multi-agent Systems (MARL): Instead of a single bot that considers the market to be a static environment, this approach creates a simulation of a market populated by multiple competing bots. This is a much more realistic model, similar to a poker game., where it is necessary to take into account the actions of other players.[36] [37] [38] [39] [40]
• AI in DeFi (DeFAI): Imagine autonomous AI agents living right in the blockchain, who themselves trade, manage liquidity, and even vote in the DAO. This is the end point of the AI and crypt merger. — a fully autonomous financial system.[41] [42] [43] [44] [45] [46]

I have long come to the conclusion that the race for alpha will lead us to hybrid systems. Just predicting the market (as CNN-LSTM does) is not enough.[9] [20] It is naive to simply make decisions in a vacuum (as a single-agent RL).[28] But to model the market as a game with other agents (MARL) — this is closer to the truth.[36] [37] The winner will be the one who can combine all three approaches. Imagine a system: the predictive model generates possible scenarios, the RL agent chooses the best move, and the MARL simulator loses how competitors will react to this move. This is a forecast-action-simulation cycle, and in my opinion, this is exactly what the next-generation alpha machines will look like.

Platform Battleground: Who’s Who in the Market

Having dealt with the technologies, let’s look at the tools that are available to ordinary users. There is a war going on here — a war for the client, and each of the major players has its own strategy.

Commercial giants: Bitsgap, 3Commas and Cryptohopper

These three companies are the heavyweights of the market. At first glance, they look similar, but the devil, as always, is in the details.

• What they have in common: All three offer a standard gentleman’s set: bots for grid trading (Grid) and averaging (DCA), smart terminals and portfolio management on various exchanges.[47] [48] [49] [50] [51]
• And here are the differences:
  • Bitsgap is like a Toyota Camry in the world of bots. Incredibly reliable, intuitive, and with an impeccable reputation for security (they say, “zero hacks”). The perfect choice for those who appreciate simplicity and restful sleep.[47] [52] [53] [54]
  • 3Commas is a “Swiss knife” with a bunch of functions. Here you will find an advanced terminal and integration with signals from TradingView. But there is one “but”, and it is huge.: their reputation was trampled after a massive leak of users’ API keys.[49] [51] [53] [55]
  • Cryptohopper has bet on the community. Its main feature is a huge marketplace where you can buy and sell ready—made strategies, signals and templates. This is a paradise for social trading and for those who want to try something exotic., like arbitration.[56] [48] [49] [50] [57]
• Issue price: They all work by subscription. Study the tariff plans carefully: in some cases they limit the number of bots, in others they force you to pay extra for futures or access to “smart” functions.[50] [58] [59] [60]

The $20 Million Lesson: The Story of 3Commas’ Failure Let’s look at this point in more detail. The incident with 3Commas at the end of 2022, when users lost more than $20 million, is not just news. It’s a cold shower for the entire industry.[61]

• What happened: Some “well-wisher” leaked 10,000 API keys of users to the network. These keys are like duplicate keys to your apartment. The attackers simply entered the exchanges on behalf of users and began to create game, pouring money into their pockets.[61] [62]
• Effects: An FBI investigation, a ruined reputation, and a very important lesson: by entrusting your keys to a cloud platform, you are entrusting your money to it.[53] [61]
• Output: This incident made many people think. He highlighted the value of platforms with a reinforced concrete reputation like Bitsgap [52], [53]and open source solutions where your keys never leave your computer.[55]

Alternative path: Open-Source

For those who do not trust the “clouds” and are ready to get their hands dirty, there is the world of open-source. Here you get full control, but all the responsibility lies with you.

• Hummingbot: This is a tool for the pros. He is trained for complex strategies like market-making and arbitrage and knows how to work with both centralized and DEX exchanges. But be prepared for the fact that you will have to write code.[63] [64] [65]
• Freqtrade: Flexible Python bot, which is loved for its powerful engine for testing strategies on history (backtesting) and the ability to screw any machine learning models. This is a real constructor for a developer.[64] [65] [66]
• Compromise: The choice here is simple: pay with money for convenience and a beautiful interface, or pay with your time and knowledge for full control and security.

Platform	Main Focus / Strategy	Target User	Key Difference	Exchange Support (CEX / DEX)	Price Model	Security Reputation (Incidents)
Bitsgap	GRID bots, user-friendliness	Beginners and experienced traders	Intuitive interface, strong security	15+ CEX	Subscription	Zero hacks since launch
3Commas	DCA bots, smart terminal	All levels	Wide range of tools, TradingView integration	12+ CEX	Subscription	Major API Key leak (2022)
Cryptohopper	Social trading, customization	Marketplace-oriented traders	Marketplace for strategies and signals	15+ CEX	Subscription	Past issues reported, improved over time
Hummingbot	Market-making, arbitrage	Developers, quantitative traders	Open-source, professional-grade strategies	19+ CEX, 24+ DEX	Free (open source)	User-controlled, no centralized incidents

Part 3: Managing Speaker for trade work

New Wave: new signals

To begin with, this is a new class of players who do not offer bots, but offer “brains” — analytics and trading signals from THEM. Example — Intellectia.ai .

• Fashion designer: He offers you as a “financial AI agent” who makes you ready people for one person.[67] [68] [69]
• Tasks: That’s why we call it a very simple process. Users complain about technical glitches and the questionable value of the signals. Classic “growth diseases”.[68]

Power over the “Grail”: Life and the Amount of Difficult Entrepreneurship

The current one is with the most important thing. About what bot sellers don’t like to talk about. About the search and the inevitable loss of a trading advantage, or, as they say in the industry, “alpha”. This is the heart of all religious trading.

From and to the meeting: the quantum trader’s kitchen

Creating a work strategy is not magic, but hard, disciplinary work. This is probably the best experiment to date.

• Idea and data: It all starts with a hypothesis. For example, “after a big drop, the price always bounces back a bit.” Then you need to find high-quality historical data to verify this.[70] [71] [72] [73]
• Backtesting: This is a proposal of your idea on a historical level. It sounds simple, but here lies the main love — retraining. This is when your model has learned the past so well that it has become completely useless for predicting the future. She learned not a pattern, but random noise.[74] [75] [76] [77]
• Strength test: To avoid falling into the trap of overfitting, it is necessary to use more sophisticated methods.
  • • Tests on “unseen” data (out of sample): Some of them are specially designed to help with learning, and some are tested on themselves. It’s like giving a student a ticket they’ve never seen.[74] [76] [77]
  Optimization “step by step” (moving forward): This is an even more rigorous method that mimics real trading. The model is optimized on one piece of history, tested on the next, and then shifted further. This is critically important to understand., is the strategy able to adapt to a changing market?[78] [79] [80] [81]
• Questions: What does it mean if the parameters don’t remember anything? If your strategy only works with a moving average with a period of 13, but breaks down at 12 or 14, you have a problem. A reliable strategy must be sustainable.[82] [83] [84] [85]

Independent neighborhood: why loving strategy is important

Perhaps you helped me work and found the right strategy, and this is important news for you. She’s going to die. This phenomenon is called scarlet decay.

• Which is exactly the case: Imagine that you have attacked this place where people are hiding. As soon as others see that you are digging there, they will come running with their shovels, and soon there will be no coins left. It’s the same in the market. As soon as the strategy becomes known, it begins to be copied, and the advantage disappears.[86] [87] [88]
• Race to the bottom: Technology has accelerated the process a bit. In the world of high frequency technology (HFT), microseconds can be used. The more players use similar technologies, the faster any “alpha” turns into a pumpkin.[87] [89] [90] And then show, that this disintegration is happening faster every year.[87]
• In search of a better life: How do I start fighting? The only way is to diversify. You need to have a whole portfolio of different, unrelated strategies. When one dies, the others continue to work. And, of course, constantly looking for new ones., not yet hackneyed ideas.[91] [92] [93] [94]

What else: when bots become adults

When thousands of bots start doing the same thing, one trader’s problems can turn into a problem for the entire market.

• Main’s staffing table: Regulators are not stupid people. If everyone uses similar models trained on the same data, then they will react the same way to the same event. For example, everyone will start selling at the same time. This creates a strategic “monoculture” and can lead to numerous flash crashes, no matter how it was in 2010.[95] [96] [97]
• Barbell principle: How to protect yourself from the unpredictable? Nassim Taleb came up with the concept of antifragility — to create systems that not only survive in chaos, but become stronger from it. In trading, this can be a “barbell strategy”: you keep most of your capital in ultra—reliable assets, and a small part in very risky ones with unlimited growth potential. This is how you are protected from “black swans”.[98] [99] [100] [101] [102]

And here we come to the main conclusion. A successful AI trading company is not one that has a single genius bot. This is the one that has a “factory” for the production of these bots. Any strategy — It’s a depreciating asset, like a car that loses value as soon as it leaves the showroom.[86] [87] Therefore, a successful business here is more like a pharmaceutical company: it must have a constant pipeline of new “drugs” (strategies) to replace those, whose “patent” has expired (alpha).[72] [73] [103] [104] So if you’re evaluating such a company, don’t ask, “What’s your profitability?” Ask: “What is your strategy creation and decommissioning process? How fast do your alphas die and what do you have in development to replace them?”. It changes everything.

The Sheriff goes to town: Regulators and Geopolitics

For a long time, AI trading existed in the “gray zone”, like a saloon in the Wild West. But that time is running out. The sheriffs, represented by regulators, are coming to the city, and a big geopolitical game for control of technology is unfolding in the background.

The End of the Wild West: regulators tighten the screws

The era of “do what you want” is coming to an end. Especially in the United States, regulators have moved from passive surveillance to active action.

• The fight against “AI-washing”: Tired of hearing how every other company attributes to itself the magical properties of “AI”? The U.S. Securities and Exchange Commission (SEC) also. In March 2024, they fined two firms for the first time for lying about using AI. It was a clear signal.: They will hit your pocket for empty marketing.[105] [106] [107] [108] [109]
• Old rules for new toys: Other agencies, like the CFTC and the NFA, have unequivocally stated: we don’t care what you call your system — “AI”, “algorithm” or “magic box”. The existing rules of supervision and risk management apply to you as well.[110] [111]
• The path to legalization: But there is some good news. In September 2025, the SEC and the CFTC finally released a joint statement that clarified how regulated exchanges can legally trade spot crypto assets. This is a huge step towards making the crypt stop being a toy for geeks and become part of a large financial system.[112] [113]

A new big game: “Sovereign AI” against the crypt

While lawyers are arguing about the rules, another game is unfolding at a higher level. Countries are beginning to view AI as a strategic resource that needs to be controlled. And this goes against the very idea of cryptocurrencies.

• What is “sovereign AI”: It’s a simple idea: a nation must control the data, algorithms, and computing power on which its economy and security depend. You can’t trust another country with this.[114] [115]
• National projects:The United States, China, the United Arab Emirates, and Canada are all investing billions in building their own data centers and supercomputers in order not to depend on other people’s technologies.[116] [117]
• What does the crypt have to do with it? And despite the fact that the whole philosophy of the crypt is decentralization and the absence of borders. And “sovereign AI” is about national control and borders. This conflict could lead to market fragmentation, demands to store data internally (which would kill many DeFi projects), and even the use of AI in government cyberattacks on financial markets.[118] [119]

As a result, we get the classic trilemma, as in the textbook. You can only choose two of the three: (1) decentralization, (2) high productivity, (3) national sovereignty. Powerful AI requires centralized computing clusters.[9] [26] [116] States want to control these clusters.[114] [115] [117] And the crypt wants to, so that there is no control at all.[120] [121] Something It will have to be sacrificed:

• You can be decentralized and sovereign, but then you will most likely lose out in performance.
• You can be productive and decentralized, but then control will be in the hands of several global IT giants, and goodbye sovereignty.
• You can be productive and sovereign (that’s what it’s all about), but then forget about decentralization.

I think the market will split. There will be a “government” camp with powerful, regulated platforms and a “guerrilla” decentralized camp that will sacrifice productivity for freedom. And this confrontation will determine the development of the market for the next ten years.

What should I do? The strategy of survival on the border of man and machine

So, we have dismantled the market by the bones. Now the main question is: how to survive in all this and, perhaps, even succeed? The answer lies not in technology, but in understanding the new “meta-game” and the changing role of man.

The “meta-game” in trading

Being the best at trading is not just about having the fastest algorithm. It’s like in poker: you not only need to know your cards, but also understand the psychology of other players at the table.

• Game on game: Success is about understanding the “meta-game.” It is necessary to analyze not only the market, but also the ecosystem of other strategies, behavioral errors of other traders and how the market structure itself is changing.[122] [123] [124]
• Capitalize on other people’s fears: Strangely enough, the most stable strategies are often based on exploiting predictable irrationality. Panic, greed, and herd feeling all create patterns that AI can learn to recognize and use.[19] [125]
• Meta-labeling: An advanced technique for paranoids. Imagine that you have one model that generates trading signals. And on top of it, the second model works, which decides whether to trust the signals of the first one. It separates idea generation and risk management, and it’s pretty damn effective.[123]

 The new role of man: from trader to overseer of AI

Do you think bots will completely replace humans? I don’t believe it. The role of a person will not disappear, it will just change. We are turning from performers into strategists and supervisors.

• Our demons stay with us: Even with bots, we manage to sneak our psychological bugs into trading. Arrogance, fear, and greed all influence how we create bots and, more importantly, when we interfere with their work. How many times have I seen a trader turn off a profitable bot in a panic at the very first drawdown![126] [127] [128]
• Man + AI = Team: The future does not lie with full automation, but with effective collaboration. A person sets a strategy, understands the context, and creates. AI is tirelessly executing, analyzing data, and maintaining discipline. It’s like a good pilot and a smart autopilot — they complement each other.[129] [130] [131] [132]
• The trap of blind trust: There is also a downside — the risk of overly trusting the machine, stopping checking its decisions and asking questions. In finance, where real money is at stake, such a “bias towards automation” can be very expensive.[133] [134]

Transparency requirement: Why do we need “explicable AI”

The more complex the models become, the less we understand how they make decisions. They turn into “black boxes”. And this is a big problem.

• Why is this important?:Explicable AI (XAI) technologies are trying to make these boxes transparent. This is necessary for three things: to debug models so that users can trust them and, what will soon become the most important thing, to comply with regulatory requirements.[135] [136] [137] [138] [139]
• Regulators will require: I’m pretty sure that future laws, like the European AI Act, will oblige financial companies to explain how their algorithms work. And those who do not prepare for this in advance will find themselves in a very unpleasant situation.[135]

So what’s the end result? Specific tips

Let’s draw a line. Here are my recommendations for those who are somehow connected with this market.

• For investors and funds:
  • Stop looking for one “grail”. Build a portfolio of different, unrelated strategies. This is the only way to survive the “alpha breakup.”
  • When evaluating companies, look not at their past returns, but at their process of creating and managing strategies. Ask about their “alpha factory”.
  • Start training your traders to be AI managers rather than performers. Their new job is to set tasks and monitor, not push buttons.
• For platform developers (Bitsgap, Cryptohopper, etc.):
  • Safety, safety, and more safety. One major hack and your business is gone.
  • Shift your marketing focus from “make a million” to “keep your discipline and nerves.” Speak to Generation Z in their language, sell not profit, but cognitive load relief.
  • Start implementing explainability functions (XAI). This will strengthen user trust and prepare you for future regulatory requirements.
• For technologists and researchers:
  • Focus on the hybrids. The future belongs to systems that can predict, make decisions, and model competitor behavior (MARL).
  • Dig towards causal AI. This could be the next big breakthrough that will provide truly sustainable models.
  • Finally, create normal, standardized tests to evaluate the reliability of algorithms. Simple backtests have long been insufficient.