In 2006, Michael Burry walked into Goldman Sachs and asked to buy credit default swaps on a specific basket of mortgage bonds. The contracts didn't really exist at scale yet. He had to negotiate them line by line. It was bespoke risk, traded bilaterally through humans on paper over weeks and months.
Twenty years later, much of the workflow for transferring custom risk remains slow and manual.
Now, imagine the same conversation, but the buyer is an AI agent and the seller is an AI agent, too. That is the world we are entering, and the infrastructure to support it does not yet exist.
<h2 id="the-agentic-economy-is-already-here">The Agentic Economy Is Already Here</h2>
There are <a href="https://blockeden.xyz/blog/2026/04/20/250k-daily-active-on-chain-ai-agents-400-percent-growth/" rel="noopener noreferrer nofollow">roughly</a> 250,000 daily active agents <a href="https://coinmarketcap.com/academy/glossary/transaction-tx" rel="noopener noreferrer nofollow">transacting</a> onchain, growing 400% year over year. In financial services, non-human identities <a href="https://www.cyberark.com/resources/financial-services/96-machines-per-human-the-financial-sectors-agentic-ai-identity-crisis" rel="noopener noreferrer nofollow">outnumber</a> humans by 96 to 1. Coinbase and Cloudflare's x402 payment standard for machine-to-machine transactions just <a href="https://sanbi.ai/blog/agent-stack-protocols-2026" rel="noopener noreferrer nofollow">crossed</a> 165 million transactions in its first nine months. None of that was touched by a human.
This economy is developing around four layers: <a href="https://coinmarketcap.com/academy/glossary/digital-identity" rel="noopener noreferrer nofollow">identity</a>, including ERC-8004; wallets, such as Coinbase Agentic; payments through x402; and execution through <a href="https://coinmarketcap.com/academy/glossary/decentralized-exchange-dex" rel="noopener noreferrer nofollow">decentralized exchanges</a>, <a href="https://coinmarketcap.com/academy/glossary/defi-lending" rel="noopener noreferrer nofollow">lending markets</a>, and vaults. The stack is largely complete.
What it still lacks are instruments that can make those structures more resilient when something goes wrong. One critical layer is missing: risk.
<h2 id="risk-has-to-be-agentic-too">Risk Has to Be Agentic, Too</h2>
Agents today can create risk and amplify risk, but they cannot meaningfully manage or transfer it. And we’ve already seen the effects of that gap.
When the LayerZero/rsETH exploit <a href="https://www.galaxy.com/insights/research/kelpdao-layerzero-exploit-defi" rel="noopener noreferrer nofollow">hit</a> earlier this year, vault curators used automated exit tools within their Morpho positions. The moment the risk signal crossed a threshold, those agents executed exactly as designed: liquidate, exit, and retrieve the capital. They moved in milliseconds to protect depositors from the worst-case outcome. That is best-in-class agentic risk management today.
But notice what was not available. The agents could not hedge or buy partial coverage against an incident like this. They couldn't negotiate a countertrade with another agent that had a different view. They could not structure a swap that paid out below a net asset value (NAV) threshold while preserving the upside.
Their only tool was out.
When every agent's only tool is out, the second-order effect is predictable: cascading exits as agents rush out of markets and <a href="https://coinmarketcap.com/academy/glossary/liquidity" rel="noopener noreferrer nofollow">liquidity</a> dries up at machine speed.
These are risks that could be priced and traded. Instead, they are currently resolved by everyone running for the door at once. This amplifies volatility and systemic risk because non-human actors lack the option to do what Burry did.
<blockquote>
 Read More: <a href="https://coinmarketcap.com/academy/article/ai-agents-tech-stack-getting-more-powerful" rel="noopener noreferrer nofollow">The AI Agent Tech Stack Is Here, and It’s Only Getting More Powerful</a>
</blockquote>
<h2 id="programmability-enables-risk-customization">Programmability Enables Risk Customization</h2>
Many hear “risk” and think of hedging, or insurance, or risk-on versus risk-off positioning. In a programmable onchain system, the more useful frame is “risk customization.”
Take a single yield-bearing vault token, for example. Now imagine three different users looking to hedge their positions. We have one asset and three different needs. In programmable finance, each could instruct a bespoke agent:
User A says: "Protect me if NAV drops more than 3%. If that happens, let me swap my vault token for stablecoin at 97 cents on the dollar."
User B says: "I don't care about downside. I care about liquidity. I want an on-demand exit at NAV, anytime."
User C says: "I want downside insurance. Pay me up to 10 cents per dollar if NAV slips below par."
These are three different risk shapes for the same underlying asset, with different triggers, settlement mechanics, and premiums. That’s a very different scenario than what we see in <a href="https://coinmarketcap.com/academy/glossary/tradfi" rel="noopener noreferrer nofollow">TradFi</a>, or even DeFi, where those products do not exist. You either take what's on the menu or you walk away.
Why doesn't this market exist already? Human underwriters cannot price countless variations of the same risk in real time. Geico solved real-time pricing in the 1990s by replacing the human with an algorithm, but insurers still offer a fixed menu. Agents could replace the menu itself.
However, none of it works on the rails we have today. Agents can't transact in TradFi infrastructure because it’s built around human counterparties, standardized derivatives agreements, <a href="https://coinmarketcap.com/academy/glossary/kyc" rel="noopener noreferrer nofollow">know-your-customer (KYC)</a> checks, and unstructured emails. Agents walk up to that infrastructure and bounce off.
What agents need are structured data, deterministic addresses, atomic settlement, and composable states. All of this exists in onchain finance.
<h2 id="risk-customization-will-be-the-norm-but-we-need-to-build-it-first">Risk Customization Will Be the Norm, but We Need to Build It First</h2>
Most crypto trading volume is already AI-driven and will benefit from these new capabilities. Traditional finance is likely to follow as it recognizes the advantages of customizable risk over today’s “take-it-or-leave-it” model.
However, that risk layer still needs to be built and adopted. Right now, DeFi has roughly $79 billion in <a href="https://coinmarketcap.com/academy/glossary/total-value-locked-tvl" rel="noopener noreferrer nofollow">total value locked (TVL)</a>. Less than $125 million of that, about 0.15%, carries any risk coverage, and almost entirely for smart contract risk. The financial risks that matter most in DeFi — like vault NAV drops, stablecoin depegs, liquid staking token (LST) slashing, and bad debt — have <a href="https://www.coindesk.com/business/2026/05/16/crypto-users-are-choosing-juicy-yields-over-protection-putting-billions-at-risk-of-hacks" rel="noopener noreferrer nofollow">effectively zero</a> coverage today.
Over the next 24 months, capital formation in agentic finance will take one of two paths: programmable risk primitives built on transparent, <a href="https://coinmarketcap.com/academy/glossary/composability-defi" rel="noopener noreferrer nofollow">composable</a>, fully collateralized, machine-native infrastructure, or the same opaque, bilateral, slow plumbing that gave us 2008.
If programmable risk markets take off, then The Big Short scene in 2028 will look nothing like the one in 2006. A user will simply type into a chat interface: “Hold this position for six months, need 100% exit liquidity, $250,000 notional.” Then an agent will find three underwriters, compare quotes, mint the position to the user's wallet, and settle atomically on acceptance.
Risk infrastructure has to be in place before that conversation can happen.
<blockquote>
 Read More: <a href="https://coinmarketcap.com/academy/article/bitcoin-miners-pivoting-ai-what-mean-btc" rel="noopener noreferrer nofollow">Bitcoin Miners Are Pivoting to AI: What Does This Mean for BTC?</a>
</blockquote>

Risk Is the Missing Layer in the AI Agent Economy

Table of Contents

Other articles published on Jul 14, 2026

Kraken's Parent Plans US Hyperliquid Perps: Will They Feel On-Chain?

Ethereum Price Stalls as DEX Trading Share Falls to Lowest Since 2018

Raydium Price Rises 64% as Tokenized Stock Trading Shifts to Solana

Join our free newsletter for daily crypto updates!