Most people still think an AI managing their money is years away. It is already happening. To understand how close we really are, we sat down with someone who works on it full-time: DimeTV guest Patrick Delaney, CEO and co-founder of Ampli, which builds the security and orchestration layer for AI agents that move real money on-chain. His core message is that agentic finance is here in an early form, and the thing standing between it and the mainstream is not smarter models. It is trust.
This isn't a how-to on trading mechanics. If you want the foundations, our options series covers calls, puts, the Greeks, and how implied volatility is priced. This is a forward look at agentic finance from a builder in the middle of it: what it is, why crypto is the backend for it, what it takes to trust an agent with capital, and where it goes next.
Watch the full episode on DimeTV with Ampli's Patrick Delaney, or read the highlights below.
Featured guest
Patrick Delaney, CEO and co-founder of Ampli, which builds security and orchestration for AI agents that autonomously manage capital on-chain: an on-chain policy engine and MPC-based wallet security that sit between the agent and the money. Trained as a political scientist, Patrick came into crypto through the governance side of blockchains and spent years advising protocols on tokenomics and go-to-market, including the agentic-finance pioneer Giza. He has presented on the future of money at the World Bank and joined Paradex on DimeTV to talk about what happens when AI agents start managing real money.
From the episode
Agentic finance is already live, not years away. Autonomous agents are managing real capital on-chain today, starting with experienced DeFi users.
Crypto is the backend for agentic money: an AI agent cannot open a bank account, but it can hold a wallet and move programmable money 24/7.
The biggest use case may not be finding alpha. It is risk management: agents that watch positions around the clock and exit danger before a human could react.
The real risk is not rogue agents, it is governance and security. Most losses last year came from human error and weak operations, not smart-contract bugs.
The unlock is infrastructure-level guardrails: separating the agent's intelligence from execution and custody, so an agent can act but cannot exceed its limits.
What is agentic finance?
Agentic finance is the use of AI agents to manage money autonomously: software that holds assets, executes transactions, manages positions, and reacts to markets without a person approving each action. Patrick's framing is blunt: this is not a future scenario, it is ongoing. At a conference a year ago, panelists kept saying agents would start handling real money in about two years, while the people who had already shipped it sat in the audience and smiled, because their agents were doing it that day.
The clearest proof point is the agentic-finance pioneer Giza, whose ARMA agent moved capital across lending markets entirely autonomously. By Giza's own published figures, the agent generated roughly $5.4 million in transaction volume from about $930,000 in deployed capital on Base in a matter of weeks, with no human approving the trades and every move serving a purpose rather than churning. Agentic finance, in other words, has already left the demo stage. What is still early is the audience: today it is mostly experienced on-chain users spinning up an agent from their own wallet, and the next leg is bringing in capital that sat out the first cycle.
Why is crypto the backend for agentic money?
There is a simple reason AI agents are learning to manage money on crypto rails first, and it is not ideology. An AI agent cannot pass the checks a bank requires to open an account. It has no legal identity, no passport, no signature. But generating a crypto wallet takes seconds, and the money that lives in it is programmable, global, and available every hour of the day. That structural fit is why Patrick describes the company's job as the backend for agentic money: the orchestration layer that lets an agent actually hold and move value safely.
He makes a second point that is easy to miss. The crypto industry spent years battle-testing these rails, often through painful events, before anyone knew AI would be the heavy user. A statistic he cites from Visa captures how ready the rails were: by 2024, a large majority of stablecoin transaction volume on Ethereum was already driven by bots rather than people. The machines, in effect, have been the better users of this infrastructure for a while. Now that the rails are hardened, they are ready for the next step, which is agents managing capital on behalf of humans.
Want to trade on rails built for what comes next? Paradex offers BTC and ETH options, perps, and onchain vaults from a single self-custodial account, with deep liquidity and an experience built to be acted on, by people and, increasingly, by agents.
The real use case isn't alpha, it's risk management
Everyone assumes agents are for chasing returns. Patrick's view is that the more durable value is on the other side of the trade: keeping you alive. The most striking fact about institutional crypto, he notes, is that the funds that win are usually the ones that simply survive, and survival is exactly what an always-on agent is good at. DeFi is a 24-hour market, and a human desk is not staffed around the clock. An agent is. It can watch the health of dozens of markets in parallel and act the moment something breaks.
He points to cases where agents pulled capital out of a destabilizing protocol before a human could react: when danger loomed, the agents took the cautious path and moved funds to safety. As he puts it, you are rarely sad about the trade you didn't take on a bad day; the value is in stepping back, assessing, and only redeploying once the situation is clear. For a fund manager, that is worth a great deal during working hours and even more during the hours they are not at the desk.
Why survival is the killer feature
For the average DeFi user, an agent mostly improves yield by being more active and seizing more opportunities. For a serious desk, the yield is already handled; what an agent adds is coverage. Markets move at 3am, and an agent that is consuming data in parallel and watching risk parameters every second is the difference between exiting a bad position in time and waking up to a loss. The first job of capital, in Patrick's telling, is not to grow, it is not to die.
Can you trust an AI agent with your money? AI agent security, explained
This is the question the whole field turns on, and it is the problem Ampli was built to solve. The instinctive fear is a rogue or hallucinating agent emptying a wallet. His counter is that the bigger danger is more boring and more common: governance and operational security. He cites figures from the security-research community showing that the large majority of crypto value lost last year came from human error, weak operations, and social engineering, not from smart-contract exploits. The threat is often a person being tricked or a sloppy setup being compromised, not the agent itself going haywire.
The architectural answer is to stop giving agents the keys. In safer designs the intelligence layer is separated from execution and custody, the approach Patrick's company brands as its Separation of Powers architecture. The agent can propose an action, but it never holds a raw private key. Instead, an on-chain policy engine and an MPC wallet, which splits the signing key into shares held by different parties, enforce hard limits on what can actually be signed: spending caps, approved contracts, and the like. Even a compromised or confused agent cannot move funds outside the rules, because it physically cannot produce a valid signature on its own.
Deterministic beats clever, where money is concerned. Patrick notes that the most reliable systems often use deterministic models for the actual trading logic and reserve large language models for research and scanning for risk. When real capital is moving, predictability matters more than improvisation, which is another layer of security on top of how the funds are signed.
What are agentic payments, and why blockchains?
Beyond managing portfolios, agents will increasingly transact with each other. Agentic payments are transactions initiated and settled by software rather than people: an agent paying for data, compute, or an API, or one agent paying another to complete a task. Patrick frames agent-to-agent exchange as a natural fit for blockchains, because it mirrors the original problem they were built to solve, letting parties who cannot fully trust each other transact safely without a middleman.
His longer-term intuition is striking: if you imagine highly capable AI systems acting in their own interest, they may prefer settling on open networks where they can run their own infrastructure rather than relying on a single company's rails. He is careful not to overstate it, much of everyday agentic commerce may simply run on upgraded traditional payment systems, but for autonomous capital management and the more complex end of the spectrum, he sees blockchains as close to inevitable.
What does agentic finance mean for traders and risk teams?
The natural worry is that agents replace the humans. Patrick's read is that the operator is not replaced, but augmented. He describes it almost in cyborg terms: the trader inherits some of the agent's superhuman capabilities, processing more data in parallel than any dashboard could show, while still setting the intent and the boundaries. Good judgment plus an agent's reach beats either one alone. The interface matters too; dropping a person into an empty chat box and asking what they want is exhausting, so the winning products keep a strong, guided experience around the agent.
This is squarely the thesis behind what Paradex is building with Dime Terminal: an agent that watches your account and acts on a trust ladder, where you decide how much autonomy to grant. For pure risk management you may want full autonomy, because an agent that has to wake you for approval will not exit the market when it should. For opening a position you may prefer a human in the loop, where the agent surfaces a good setup and you decide whether to take it. The same supervised-agent pattern, with the human setting the rules, is what both Patrick and Paradex are betting the market moves toward.
The Paradex angle: if a capable AI agent can choose where to trade, it will weigh the things that actually matter to it, execution costs and fees, liquidity, and privacy. Much of what Paradex has built is aimed at being the venue an agent would pick on those terms, and at being discoverable and connectable by agents in the first place. As Patrick put it about the partners he works with: when users come back to a protocol, increasingly it is agents acting on their behalf, so you had better be ready for them.
What is the one thing to take away?
Agentic finance is not a someday story, it is an early-stage reality, and crypto is the financial backend making it possible. The bottleneck is not intelligence, models are improving and even commoditizing; the bottleneck is trust, and trust is an infrastructure problem. The systems that win will be the ones that let an agent act with real autonomy while making it impossible for that agent to step outside its limits. For traders, the practical takeaway from a builder in the middle of it is that the tools are arriving to express a view and have it executed and risk-managed for you, and the question worth asking now is not whether agents will manage money, but whether the rails you use are built for them.
Frequently asked questions
Agentic finance is the use of AI agents to manage money autonomously: software that can hold assets, execute transactions, manage positions, and react to markets without a human approving each action. According to Ampli's Patrick Delaney, it is already happening rather than years away, starting with experienced on-chain users and now moving toward institutional capital. Crypto is the natural home for it because an AI agent cannot open a bank account, but it can hold a wallet and move programmable money around the clock.
No. A traditional trading bot follows fixed, preset rules and tends to break when conditions change. Agentic finance describes AI agents that can reason, adapt, and act across protocols, monitoring markets, managing risk, and executing strategies, often with a human setting the boundaries rather than approving every step. Many of the most reliable systems use deterministic models for the trading logic and keep large language models for research and risk scanning, precisely because predictability matters when real money is moving.
The core of AI agent security is never giving an agent a raw private key. Safer designs separate the intelligence layer from execution and custody: the agent proposes an action, but a policy engine and an MPC wallet, which splits the signing key into shares held by different parties, enforce hard limits on what can actually be signed. According to Ampli's Patrick Delaney, most crypto losses last year came from human error, weak operational security, and social engineering rather than smart-contract bugs, which is why infrastructure-level guardrails, not just trust in the agent, are what make autonomous capital management defensible.
Agentic payments are transactions initiated and settled by autonomous AI agents rather than people: an agent paying for data, compute, or services, or one agent paying another. Blockchains suit this well because agents need money that is programmable, global, and available 24 hours a day, which traditional banking rails were not designed to provide. It is an emerging area, and the security and policy controls around it are still being built out.
Yes, though still in an early-adopter phase. Autonomous agents are already deployed on-chain. Giza's ARMA agent, for example, generated roughly 5.4 million dollars in transaction volume from about 930,000 dollars in deployed capital on Base in a matter of weeks, with no human approving the trades. The next phase, according to Ampli's Patrick Delaney, depends less on smarter models and more on the security infrastructure that lets institutions and everyday users trust an agent with their capital.
About DimeTV
DimeTV is the media channel of Paradex, where the operators, traders, and builders shaping crypto's market structure share what they're seeing. This episode features Patrick Delaney of Ampli. Thanks to Patrick for joining us. Follow DimeTV on X for new episodes.
This content is for informational purposes only and does not constitute financial, investment, or trading advice. The views expressed are those of the guest and do not necessarily reflect those of Paradex. Trading options, perpetuals, and other crypto derivatives involves substantial risk, including the loss of premiums paid and, on sold positions, losses beyond the premium received. Leverage can amplify losses, including the rapid and total loss of capital during volatile events. Past performance is not indicative of future results. Do your own research before trading.
References to agentic finance, AI agents managing capital, agentic payments, and vaults describe an emerging and evolving area, including features and products that may be on a roadmap rather than currently available. Autonomous systems carry their own operational, security, and smart-contract risks. Availability of products varies and some may not be live. Access to Paradex may be restricted in certain jurisdictions. Verify your local regulations before using the platform.
Quotes and paraphrased remarks are drawn from the DimeTV episode and have been lightly edited for clarity and length. Third-party figures, including those attributed to Giza and Visa, reflect the guest's references and published sources, not Paradex data.
Most people think an AI managing their money is years away. It is already happening. Ampli co-founder Patrick Delaney joined DimeTV to explain agentic finance: AI agents that hold and move real money on-chain, why the real use case is risk management and not alpha, and what AI agent security has to get right before you trust an agent with capital.
.svg)
