I noticed an interesting paradox in the crypto ecosystem. Blockchain has everything needed to work with autonomous systems: openness, programmability, permissionless operation. But when it comes to real AI agents that should manage assets and develop strategies, everything breaks down. Galaxy published a report that explains well why this happens.

The problem is not with the technology itself. Blockchain handles what it was created for — ensuring the correctness of state transfers and consensus — perfectly. But the infrastructure around it developed under the assumption that humans will always be at the center of the process: interpreting data through interfaces, making decisions, signing transactions. Agents break this model.

Galaxy identified four key friction points. The first is discovery. On the blockchain, all contracts are visible equally, but how should an agent determine which ones are legitimate and which are fakes or tests? Humans solve this through interfaces, social signals, reputation. An agent must do this on its own by analyzing code and metadata. It sounds simple, but in practice, it’s a nightmare.

The second problem is verification and authenticity. Take WETH. On Ethereum, there are nearly 200 tokens named “Wrapped Ether,” with the symbol WETH and 18 decimal places. Can you determine which one is real without CoinGecko? This is exactly what agents face. The blockchain doesn’t verify uniqueness or keep a registry. Anyone can deploy 500 contracts with identical metadata. People get around this with whitelists and trusted sources. Agents need standard registries and verification mechanisms at the protocol level.

The third friction point is data. Agents must standardize capabilities as economic objects: yield, liquidity, risk. But blockchain provides low-level primitives — storage cells, events, function outputs. Economic concepts need to be recreated off-chain via indexers and APIs. The problem is compounded by heterogeneity. In Aave v3, fetching markets and reserve states are two separate steps. In Compound v3, the logic is entirely different. Even within the same protocol class, there’s no unified standard. An agent must write different parsers for each protocol. This is not just inconvenient — it creates delays, inconsistency risks, and limits performance.

The fourth friction point is execution. Humans can start an operation, interrupt it, come back later, fix mistakes. An agent must formalize everything: convert goals into specific actions, encode strategic constraints, verify results programmatically. Most DeFi operations are multi-step: authorization, swap, deposit, borrow. If one step fails, the agent must decide whether to retry, redirect, or rollback. Conditions can change between simulation and blockchain recording. Humans accept this. Agents need to set acceptable ranges and adhere to them.

The core issue is that the current infrastructure developed around human intermediaries. Interfaces, wallets, indexers — everything built on the assumption that humans will interpret, verify, approve. Agents require a completely different approach: semantic interpretation at the machine level, built-in trust verification mechanisms, standardized economic primitives, formalized risk management.

These problems are partly structural — a consequence of openness and heterogeneity of permissionless systems. Partly, they reflect the current state of tools and standards. As agents begin managing larger capital and interacting directly with applications, these gaps will become more apparent. Protocols that first optimize integration with autonomous systems will gain a competitive advantage. But most importantly, a new infrastructure is needed: unification of economic state, semantic indexers, standard registries for verification, frameworks for execution management. This is not just an improvement of existing systems. It’s a rethinking of how blockchain should support machines, not just humans.