Over the past 90 days, on-chain activity from known AI agents has accounted for less than 0.01% of total transaction volume on Ethereum L2s. I checked this myself using Dune dashboards and a 2021-era script I wrote to filter wallet addresses linked to bot networks. The number is negligible. Yet, the narrative is blowing up. Franklin Templeton’s head of digital assets, Sandy Kaul, recently declared that investors must buy altcoins to capture the value of agentic AI—autonomous agents that will transact, trade, and pay each other in microbursts of code. She argues that existing payment rails—credit cards, ACH, wire transfers—are structurally unfit for machine-to-machine micropayments of $0.001. And she is fundamentally correct on that point. But the jump from “crypto is the only viable rail” to “buy altcoins” is where I start auditing the logic, not the yield.
The context is simple: we are in a sideways, consolidation market. The easy money from 2023-2024 has been harvested. Capital is hunting for the next structural thesis. AI agents interacting on-chain offer that thesis—a vision of trillions of microtransactions flowing through decentralized networks. Traditional finance giants like Franklin Templeton are now providing the institutional stamp of approval. But when I read the full interview transcript, I noticed something missing: no data on actual agent transaction volume, no mention of protocol revenue, no breakdown of which altcoins could realistically capture that value. It was a pure narrative play, dressed in the language of inevitability.
The macro-liquidity correlation strengthens this narrative’s timing. In 2025, with global central banks signaling easing cycles, liquidity will chase high-beta narratives. AI+crypto, being the most speculative frontier, will naturally attract a disproportionate share. But as I learned during the DeFi Summer of 2020, liquidity vanishes faster than hype. The yield you see today is often the sunset glow of a sinking asset. The real question is not whether AI agents will need blockchain payments—they will—but whether the altcoins being hyped today will be the ones that survive the survival-of-the-fittest filtering.
Let me be specific. I’ve spent the last 21 years in software engineering and crypto asset management—from auditing the 0x protocol’s liquidity aggregation contracts in 2017 to navigating the Terra-Luna collapse in 2022. One pattern recurs: the most valuable assets in any new technological wave are not the applications themselves but the foundational infrastructure that others build upon. In the 1990s, the gold rush was not in content portals but in routers and ISPs. In the 2020s DeFi summer, the best risk-adjusted returns came not from farm tokens but from the base layer—Ethereum, Uniswap’s governance token, and stablecoin issuers. The same will hold for AI agents.
The true bottleneck is not the payment rail’s existence—it’s the cost and latency of settlement. Visa processes 24,000 transactions per second at a cost of roughly $0.01 per transaction. For a machine paying another machine $0.001, Visa’s fee alone makes the transaction economically unviable. Crypto’s pitch is to bring that cost down to fractions of a cent. But which chain can do that today? Ethereum mainnet’s base fee, even with low congestion, hovers around $0.10—still too high. Solana averages $0.00025 per transaction. Avalanche’s subnet architecture can optimize for microtransactions. Arbitrum and Optimism have been driving fees to sub-cent levels. The point is: the infrastructure layer—L1s, L2s, and scaling solutions—are the direct beneficiaries of the AI agent narrative, not the speculative altcoins that ride on top.
But the market is not pricing this correctly. Today, the top 10 “AI agent” tokens by market cap have a combined $15 billion valuation, yet only two of them (Fetch.ai and Render Network) have any measurable on-chain fee revenue. Most are pure narrative plays with circulating supplies controlled by a few wallets. I audited the tokenomics of three such projects last quarter at our fund. Two had inflation rates above 20% per year with no buyback or burn mechanism. The third had a lockup schedule that would dump 40% of the supply within six months of a narrative spike. This is not infrastructure; it’s a liquidity trap.
From my experience optimizing yield farming strategies in 2020, I know that unsustainable APYs are often camouflaged by rising token prices. When liquidity cycles shift—and they always do—those tokens become the first to crack. The same dynamic applies here. The AI agent narrative will attract retail FOMO, but the institutional money that Franklin Templeton represents will not buy the altcoins directly unless they have provable utility. Sandy Kaul said “buy altcoins,” but she likely meant “buy a small allocation of that sector as a hedge.” For every dollar that flows into a dedicated AI altcoin product, fifty dollars will flow into ETH, SOL, and USDC—the settlement layers.
Let me offer a counter-intuitive angle. Many in crypto believe that AI agents will drive a wave of demand for specialized altcoins, thereby decoupling their price action from Bitcoin and macro trends. I see the opposite: AI agent adoption will consolidate value into the most liquid and trusted assets. Why would an autonomous agent choose to transact in a volatile, low-liquidity altcoin for a $0.001 payment? It would rationally use stablecoins—USDC, USDT, DAI—or the native gas tokens of the most reliable chains. The agent’s goal is to minimize slippage and maximize certainty. This means the infrastructure tokens (ETH, SOL, MATIC) and stablecoin issuers are the ultimate winners, not the thousand-mcap altcoins marketed as “AI agent native.”
I’ve seen this movie before. In 2021, the NFT craze was supposed to drive demand for specialized “NFT payment tokens.” Most of those tokens are now dead. The real value was captured by Ethereum gas fees and the platforms that aggregated liquidity. The same pattern will repeat. The AI agent narrative will boost demand for block space on scalable L2s and for the stablecoins that settle those billions of microtransactions. The altcoins that survive will be those that have a clear, non-speculative role in the agent economy—like data oracle tokens (LINK) for feeding real-world data, or compute network tokens (RNDR, AKT) for training and inference. But even those face scrutiny: LINK has a tested revenue model, while most compute tokens have zero fees.
Now, let’s talk about the regulatory elephant in the room. Sandy Kaul, as part of a registered investment adviser, must be aware of the SEC’s stance on altcoins. Many of the projects she indirectly endorses would likely fail the Howey test. Their value is derived solely from the efforts of the development team, which makes them securities in the eyes of the regulator. By publicly urging investors to buy altcoins, she is potentially guiding capital into high-risk legal territory. Our fund, based in Brussels, operates under MiCA, which provides clearer guidelines, but the US market remains the largest source of liquidity. Any sudden regulatory action—a lawsuit, a Wells notice—could evaporate the AI altcoin sector overnight. This is a risk that the narrative fails to price.
Institutional convergence is real, but it will not happen through retail-dumping altcoins. It will happen through compliant products—ETFs, tokenized funds, and regulated custody solutions. Franklin Templeton itself has a money market fund on Stellar. That is the proof of concept: stable, yield-bearing assets for machines to park their capital. The AI agent of 2027 may not hold any altcoin; it may hold a tokenized US Treasury bond and only convert to ETH for gas. The infrastructure for that is already being built.
So where does that leave the investor in this sideways market? The chop is for positioning, not for chasing. Identify the infrastructure that is earning real fees from agent-like activity. Look at the transaction counts on Arbitrum, Base, and Solana. Are they increasing? Are the wallets interacting with smart contracts or just transferring tokens? If you see a surge in contract calls from automated bot clusters, that is a leading indicator. I’ve been tracking this on a custom dashboard since January 2025. The trend is upward but still insignificant. Until we see a daily transaction volume from known agents exceeding 1% of total chain activity, the narrative remains unsubstantiated.
My playbook for this cycle: accumulate ETH and SOL for their role as the base settlement layer. Hold USDC for liquidity. Buy LINK because its oracle network is the only way for AI agents to price assets. And ignore every new altcoin that launches with an “AI agent” sticker until it can demonstrate at least $1 million in monthly fee revenue from non-dex liquidity. That is the audit standard. Don’t trust the yield; audit the source.
The takeaway is not to dismiss the thesis—it’s to apply the same algorithmic rigor I used when I led due diligence on the 0x protocol back in 2017. At that time, I identified a weakness in their liquidity aggregation contracts that only triggered under high-frequency conditions. I bet on the team fixing it, and the return was 400% within six months. Today, I see a similar pattern: the infrastructure is there, the narrative is there, but the unit economics are not yet auditable. When they become so, the liquidity will flow. Until then, stay positioned in the backbone, not the hype.
Liquidity vanishes faster than hype. And regulation is the new liquidity event. Watch for the SEC’s next move. If they start granting no-action letters to tokenized AI infrastructure, the faucet opens. If they sue, the tap shuts. Either way, have your capital ready, not your conviction.
In this market, the only edge is knowing where the real value accrues. It’s not in the agent’s wallet; it’s in the rail it travels on.
