How AI Is Reshaping Crypto Market Making – And Why I’m Betting On It

I’ve been working in crypto long enough to witness its wild cycles – from DeFi summers to NFT winters – and if there’s one constant, it’s that the infrastructure keeps getting smarter. Recently, I’ve become convinced that the next big leap won’t come from a new L1, or even a regulatory breakthrough. It’s going to come from artificial intelligence – specifically, how AI will transform the way we do market making in crypto.

Let me explain why I believe this is happening right now.

The Market Maker’s Dilemma

Market makers are the invisible gears of any liquid market. We quote buy and sell prices, manage our inventory, and try to earn the spread, while constantly adapting to volatility, fragmented liquidity, and high-frequency adversaries. In crypto, that job is ten times harder.

For years, the best players in the space have built massive infrastructure: algorithmic systems, smart hedging strategies, and custom tools to manage risk across dozens of venues. I’ve either built or collaborated with teams who’ve done this. But no matter how good your system is, if it’s rule-based, it eventually hits a ceiling. Markets evolve. Other bots catch up. What worked last quarter breaks today.

That’s where AI comes in – and why I’ve started to lean in hard.

Why AI is a Game Changer

Unlike traditional quant strategies, AI doesn’t need hard-coded logic to thrive. It learns. It adapts. It finds patterns I wouldn’t have spotted after weeks of modeling. And it keeps getting better the more I feed it.

1. Reinforcement Learning for Smarter Execution

I’ve been experimenting with reinforcement learning agents trained in custom crypto environments. The difference is night and day. Instead of manually adjusting my quoting behavior, the agent adapts on the fly, tightening spreads in calm conditions, skewing aggressively when volume spikes, and managing inventory to stay risk-neutral.

These agents can simulate millions of trades before going live, optimizing not just for immediate PnL but long-term survival. They’re especially strong on DEXs, where slippage and MEV attacks make static strategies fragile.

2. AI Models That Actually Predict

I used to roll my eyes at anyone claiming “price prediction.” But now, I’m training transformer-based models on multi-modal data: Twitter sentiment, GitHub commits, macro indicators, chain-specific metrics. I’m not looking for a crystal ball – I’m looking for better signals.

And honestly, it works. These models don’t predict prices precisely, but they do improve my positioning and timing. They help my market maker know when to lean in, when to sit out, and how to skew inventory dynamically.

3. Generative AI as a Strategy Co-Pilot

One of the most surprising unlocks came when I started using generative AI to simulate strategies. I’d describe a trading environment and constraints, and the model would suggest configurations I hadn’t considered – some crazy, some brilliant.

It doesn’t replace a human quant, but it acts like a sparring partner. I iterate faster, explore weirder ideas, and run backtests across extreme market regimes. It’s the most fun I’ve had designing trading systems in years.

4. Fully Autonomous Agents Are Coming

The goal isn’t to make myself obsolete, but to build agents that can operate across CEXs, DEXs, L3 order books, and aggregators with minimal intervention. These agents don’t just execute—they learn, hedge, route, and adapt in real-time.

Imagine an AI-native market maker that:

• Monitors a dozen venues simultaneously
• Routes liquidity with latency awareness
• Adapts quoting strategies to venue-specific rules
• Hedges dynamically using perps, options, or bridge liquidity

That’s not sci-fi anymore. It’s what I’m building toward.

The Tools Are Finally Here

Why now? Because the stack has matured. We’ve got:

• Rich on-chain data, thanks to subgraphs and state channels.
• Composable infrastructure, where trading agents can plug into lending, perps, options, and bridges without building it all from scratch.
• Decentralized execution networks like Yellow Network and intents-based protocols, making it easier for AI agents to actually get trades done.
• Open simulation environments, so I can train agents before risking capital.
  

It’s never been a better time to build AI-native trading systems.

But It’s Not Without Risks

There are real concerns. If we’re not careful, AI agents could manipulate markets, overfit to noisy data, or go rogue due to poor incentive design. Explainability is hard. Monitoring is even harder. And the more we automate, the more we need safeguards.

That’s why I’m approaching this with caution. I audit everything. I set kill switches. I test under extreme conditions. And I think hard about the ethical design of these systems. 

I genuinely believe AI will eat the crypto trade stack. Not all at once, and not without friction, but it’s already happening. And if you’re a founder, a builder, or just a curious degenerate, now’s the time to lean in.

This is the edge.

We’re going from scripts to agents. From heuristics to learners. From reactive trading to adaptive intelligence.

If that doesn’t excite you, you’re in the wrong industry.