‍

If you're building onchain agents right now, it seems like you have everything you need.

The agentic finance stack is maturing quickly with primitives for agents to hold assets, be identified, send payments, execute DeFi strategies, and be built quickly. In fact, there are over 250,000 daily active onchain agents running on this stack today and some estimates show 80% of crypto trading volume is already AI-driven.

‍

But the truth is there’s still a lot of work to be done…

‍

What Agentic Finance Has Today

‍

It’s worth being precise about what has actually been built before discussing what hasn't.

‍

• Wallets: Coinbase Agentic Wallets, launched February 2026, give agents non-custodial key management in TEEs, gasless transactions on Base, and programmable spending caps. Likewise, EIP-7702 is live in Ethereum's Pectra upgrade which lets standard EOAs adopt smart contract logic temporarily and enabling session key delegation for agents. Fireblocks, Turnkey, and Privy offer the institutional-grade equivalent. So the wallet layer is solved.

‍

• Payments. The x402 protocol embeds USDC payments directly into HTTP requests. Your agent can call an API, pay the paywall mid-request, and continue operating without an account. The Linux Foundation now houses the x402 standard while Google, AWS, Visa, and Microsoft are institutional backers. Coinbase's Agentic.market, live since April 2026, is the public marketplace for x402-priced agent services which have already processed 75 million transactions.

‍

• DeFi execution: Many major protocols are adapting to having agents as users. Uniswap Labs shipped seven open-source agent skills for swaps, liquidity, and v4 hook deployment in February 2026. PancakeSwap's agent toolkit covers swap planning, liquidity positioning, and farming across eight chains. Morpho partnered with Lit Protocol to enable agent lending and borrowing with encrypted signing. Your agent can be a DeFi power user on day one.

‍

• Frameworks: OpenAI Agents SDK, ElizaOS, LangChain, LangGraph, and Almanak's Python SDK are all production-grade. Wayfinder deploys agents across Ethereum, Solana, and Base via text message and Griffain even lets agents handle complex DeFi workflows through natural language. The toolchain is mature enough that time-to-deployment is measured in minutes, not months.

‍

What Is Still Being Figured Out

‍

These are all key parts of the stack but several foundational layers are earlier in their development than the narrative suggests.

‍

• Identity and consent. ERC-8004 provides a registry for agent identity with an NFT-based Identity record, a Reputation registry, and a Validation record. Over 24,000 agents have registered. KYA frameworks from vendors like AstraSync and AgentFacts are positioning themselves as the compliance layer for institutions granting agents capital authority. But identity and consent for agents is a harder problem than any of these standards have fully solved. Who authorized this agent to act? On whose behalf is it operating? What is it permitted to do, and what is out of scope? When an agent takes an action that causes a loss, which principal is accountable? These questions sit underneath every agent deployment and none of them have clean answers yet. ERC-8004 establishes that an agent exists and has a history. It does not establish what that agent is authorized to do or who bears responsibility when it does something it shouldn't. The consent and accountability layer is still being designed.

‍

• Security. The attack surface for agents is genuinely novel and genuinely unsolved. Memory poisoning, indirect prompt injection, tool-misuse exploits, and signing authority abuse are all live attack vectors that operate above the smart contract layer, which means they are invisible to audits and formal verification. OWASP has published a GenAI Data Security Risks framework. OpenClaw has proposed a five-layer agent security architecture. These are early frameworks for an attack surface that is actively being exploited. In Q1 2026, AI trading agents lost over $45M to this class of attacks. The security layer is not a solved problem yet and it is an open research and engineering challenge.

‍

• Regulatory clarity. KYA as a compliance concept is emerging, but no regulatory framework has been finalized for agents as financial market participants. When an autonomous agent causes a loss, liability has not been legally established. The SEC and CFTC have issued signals such as a joint taxonomy classifying digital commodities, SEC Chair Atkins addressing tokenization, but the specific question of agent accountability in financial markets remains unanswered. Institutions that want to deploy agents with real capital are operating without a clear regulatory floor.

‍

• Interoperability. Most agent infrastructure today is chain-specific or framework-specific. ERC-7683 cross-chain intents are a step toward agents that can operate fluidly across chains, but multi-chain agent orchestration at scale remains more a roadmap item than a production capability for most teams.

‍

The honest picture of the agentic finance stack in mid-2026 is a set of functional but incomplete layers, some further along than others, with several foundational questions still open.

‍

Why Risk Is the Most Consequential Gap Right Now

‍

There’s one other major gap in the stack today and that is a risk layer. This one is especially meaningful given that the stack has multiple open problems that users need to look out for.

‍

Agents are already moving significant capital. $28 trillion in stablecoin volume was processed in Q1 2026, with 76% of it bot and agent-driven. ARMA-style yield agents are operating across billions in DeFi TVL. Vault managers are using agents to rebalance positions in real time. The capital is onchain, the agents are running, and the positions are open.

‍

But even if there’s a lot they can do, what they cannot do is take on meaningful risk exposure without any mechanism to manage it. There is no function an agent can call to hedge a position, transfer tail risk, or buy protection against a strategy failing. Every other executable action in DeFi has a corresponding smart contract interface. Risk management does not.

‍

The $286M Drift Protocol exploit on April 1, 2026 happened because agents compounded losses at machine speed with no risk controls underneath them. The $45M in Q1 2026 agent losses followed the same pattern. Capital exposure without risk infrastructure is the gap that is causing losses right now.

‍

The Primitive That Enables Agentic Risk Management

‍

The spec for what builders actually need is actually straightforward here. They need a callable smart contract function that accepts a pegged asset and issues two composable tokens: one that lets the holder redeem at parity if a covered event occurs, and one that takes the underwriting side in exchange for yield. 

‍

That is Cork's programmable risk primitive. It’s a protocol where Cork Swap Tokens give the holder protection while Cork Principal Tokens take the underwriting side and pool operations are callable smart contract functions.

‍

Cork is building the layer that closes the risk gap for agentic finance.

‍