In the 2000s, we stopped going into bank branches and started banking online. In 10 years time, do you think we will still be managing our finances this way?
I predict today’s online financial tools will soon look archaic as we increasingly manage our finances via agents – intelligent bots that conduct financial operations on our behalf.
Luke Saunders is a Partner at Delphi Ventures and CTO at Delphi Labs.
That may sound far-fetched, but agents are already here. A liquidation bot is an agent. An arbitrage bot is an agent. The person who created it can chill on their sofa while the agent runs in the background making them money. They are niche but there are thousands of such things performing on-chain actions today on behalf of others.
Bots/agents are not people. The world of aesthetically pleasing apps with nice buttons to click does not work for them. TradFi provides APIs, but these are often half baked, closed-source, with limited documentation and inconsistent with each other. It’s very difficult for an agent (or anyone honestly) to figure out how to use them.
As I wrote in a recent post on X, while crypto UX is cumbersome for humans, its transparent, open-source, and programmable nature is absolutely perfect for agents. Code is a first class citizen here. Everything an agent needs to figure out how to achieve a goal — the source code and the data — is public. Half-baked TradFi APIs can’t compete with this. The best agents will be built on the best platform and there is no better platform than crypto.
This isn’t our reality yet, but there is a path to get us there. Let’s jump back to the agents that exist today and explore how we get from here to there.
The agents mentioned earlier (liquidation bots, etc) meet the definition of an agent, being autonomous software entities, but they are still very basic. They are usually coded specifically for a single, narrow task and without any ChatGPT-like interface to receive instructions and deliver feedback.
Big firms like Jump and GSR are able to build more complex systems able to autonomously move money around and handle problems to some extent. But these companies have a lot of money and big teams of devs. It’s not so feasible for the little guy.
Crypto sometimes feels like a solution searching for a problem, but maybe it just happens to be the perfect financial system for the machines
Technology always progresses to lower the barrier, though. What previously took a computer the size of a house and a team of scientists is later achievable by a teenager and a phone.
As AI becomes more prevalent, the nuts and bolts of which smart contract functions to call, or comprehending the data within the contract’s state, can be abstracted away. This leaves the creator/owner able to give much higher level instructions while the agent figures out how to execute them.
We can see today that LLMs like GPT-4 are already capable of understanding the intent behind a user’s message, and reason intelligently about how to respond, while at the same time having read all of the text on the internet to inform its answer. So it’s not a big leap to imagine a text-based interface to an agent that has read all relevant crypto media, ingested and understood all the data from all the blockchains, AND has the ability to execute actions on-chain.
Think about what people in crypto are spending all their time doing today:
The airdrop hunter who is spending all day finding promising new projects which seem like they might do an airdrop, guessing which actions might be rewarded and then doing them. An agent can do all that.
The meme hunter who is following some internal algo for finding new meme-coins with signs of cultish community, then buying them, then selling when interest seems to wane. An agent can do all that.
But more importantly consider how the new AI agent interface paradigm simplifies the basics of finance:
You: “I have 100,000 USDC. Get me the best low risk yield you can find. I need instant access to it so don’t lock it up for more than a week.”
Agent: “No problem. I recommend an initial distribution of 50% in Aave, 20% in Ethena, 20% in X and 10% in Y. I will monitor rates and watch for changes in their risk levels, rebalancing as necessary.”
You: “Sounds good, do it.”
People wonder how MetaMask will ever get unseated, or how frontends can be decentralized. Perhaps the answer is in the future we don’t need frontends anyway, and MetaMask will be replaced by your in-agent wallet.
TradFi agents will exist, but agents built on crypto will be better. And that’s why this is important: if agents become more and more common as I expect they will, and crypto agents offer a better experience for users, then this drives more and more users, activity and value on-chain.
Crypto sometimes feels like a solution searching for a problem, but maybe it just happens to be the perfect financial system for the machines.
Note: The views expressed in this column are those of the author and do not necessarily reflect those of CoinDesk, Inc. or its owners and affiliates.
