Agentic AI Won't Save Ethereum (But the Narrative Might)

Franklin Templeton’s research partner, Sandy Kaul, just told a Bloomberg audience that agentic AI will need blockchain payments because AI agents can’t open bank accounts. The conclusion, as she framed it, is simple: you should buy cryptocurrencies—specifically Ethereum—as a key portfolio holding. At the same time, an IMF report projects agentic AI could trigger a multi-trillion-dollar commercial shift by 2030. Ethereum had already bounced 27% from its local low to $1,930. The pieces fit together like a jigsaw. But when I look at the code behind the narrative, I see something else: a repackaging of old hopes under a shiny new label. History rhymes, but the code doesn't—and the code is what ultimately settles value. Let me step back. Agentic AI refers to autonomous systems that can negotiate, trade, execute transactions, and manage resources without human intervention. The IMF paper acknowledges that these agents will need financial rails that are programmable, global, and permissionless. Traditional banking KYC requires a legal entity. An AI agent has no passport. So blockchain, with its pseudonymous accounts and smart contracts, emerges as the natural settlement layer. Ethereum, as the largest smart contract platform with the deepest developer ecosystem and institutional trust, becomes the default candidate. That is the core thesis. But I’ve spent the last four years digging into the infrastructure layer of crypto, and I’ve learned to distinguish between narrative resonance and structural necessity. The claim that agentic AI "needs" Ethereum specifically, not just any programmable blockchain, rests on three pillars: developer mindshare, L2 scaling for low fees, and institutional comfort. All three are real. Ethereum has about 55% of DeFi TVL, the largest community of Solidity developers, and a regulatory track record that makes it the least risky choice for a Franklin Templeton. The IMF report even notes that industry participants are already racing to build these experiments on Ethereum. Yet the execution path is riddled with friction that the narrative glosses over. AI agents operate at machine speed, executing thousands of microtransactions per second. Ethereum’s L1 processes about 15 transactions per second. L2s like Arbitrum and Base push that to several thousand, but they introduce their own centralization vectors: sequencers are single points of failure, and cross-L2 liquidity fragmentation means an agent might need to hold six different tokens to pay for gas across different rollups. Every time an AI agent bridges assets, it incurs latency and slippage. The code doesn’t yet rhyme with the vision. During a 2024 audit I conducted for an AI-powered trading system, I observed the agents were explicitly designed to avoid on-chain settlement unless absolutely necessary. They used off-chain order books and only settled net positions on L2s every few hours. The reason was simple: even at L2 gas prices, the cost of a million microtransactions would eclipse the profit margin of the trades. The agents weren’t using Ethereum because it was cheap—they were using it because their creators wanted the immutability guarantee. But for pure payments, immutability is not always a feature; reversibility and speed matter more. The IMF’s multi-trillion-dollar figure is a twenty-year extrapolation, not a near-term trigger. Now, the contrarian angle: the narrative that agentic AI will drive Ethereum demand might actually be a distraction from real competitive threats. Solana, for example, offers sub-penny fees and sub-second finality, which fit the microtransaction profile much better. Several Solana-based projects are already testing agent-to-agent payment channels using compressed NFTs and state-expansion techniques. Avalanche has subnets that can be tailored for high-frequency agent settlement. Even Bitcoin’s Lightning Network, though not Turing-complete, can handle millions of small payments per second. The better question is not whether Ethereum will be used, but whether it will be the preferred settlement layer once agents begin optimizing for cost and speed. There is also a deeper economic issue. The article recommends buying ETH because agents will need to pay gas fees. But agents can pay fees using any ERC-20 token—USDC, DAI, or even project-specific tokens. In fact, for an agent that operates on a dollar-cost basis, using a stablecoin eliminates price volatility risk. If the agent has to pay gas in ETH, it must hold ETH, which introduces currency risk. The agent’s optimization algorithm might prefer to batch gas payments via a relayer network that accepts USDC and pays ETH on the backend. That weakens the direct demand link from agent activity to ETH price. The value capture is indirect and leaky. I saw this dynamic play out during the 2021 NFT minting craze. Everyone assumed that high minting activity would drive ETH demand permanently. What happened instead was that marketplaces like OpenSea introduced their own settlement tokens and relayer schemes, reducing the gas burden on users. The same pattern will repeat: agents will abstract away gas costs, using meta-transactions or "gas stations" that pool fees. The protocol that wins will be the one that offers the lowest total cost of settlement, not the one with the strongest brand. Franklin Templeton’s endorsement is powerful, but it is also a canary in the coal mine. When a traditional asset manager starts publicly encouraging retail to buy a volatile crypto asset on the back of a speculative narrative, I grow wary. The firm may be positioning itself to launch an Ethereum ETF that benefits from the hype. The IMF report is a research document, not a regulatory green light. And the 27% price bounce that preceded the interview suggests the market had already discounted some of this good news. Let’s talk about the elephant in the room: competition from other layers and from stablecoins. Stablecoins already process trillions of dollars in payments annually, and most of those transactions happen on Ethereum or its L2s. But the value accrues to the stablecoin issuers (Tether, Circle) and to the validators of the underlying chain, not necessarily to ETH holders. If agentic AI uses USDC on Ethereum, the demand for ETH comes only from gas fees, which are a tiny fraction of the transaction volume. At current L2 fee levels, a $10 billion daily AI payment flow would generate roughly $2 million in gas fees per day—around $700 million annually. That is a rounding error compared to Ethereum’s $250 billion market cap. The narrative makes ETH sound like a leveraged bet on AI commerce, but the actual leverage is low. So what is the real opportunity? It lies not in buying the narrative, but in identifying the infrastructure that will enable agentic AI payments efficiently. That might be a specific L2 that wins the "agent rollup" race, or a de-facto standard for cross-agent stablecoin settlement, or a protocol that provides trust-minimized off-chain computation with on-chain settlement. The token that captures true economic value will be the one that becomes the settlement asset for these networks, not the one that merely serves as a commodity fuel. For now, the narrative is the product. It drives attention, which drives price. But the code—the actual architecture of how agents will transact—does not yet rhyme with the story. History rhymes only when the underlying mechanics align. In this case, the mechanics are still being invented. As a research partner who has watched narratives inflate and deflate for nearly a decade, I suggest treating this as a trading signal, not a thesis. Monitor L2 transaction volumes, catch institutional ETF flows, and keep an ear on what the builders are actually shipping. The truth settles in the blocks, not in the headlines.