Skip to content

The CCS Interview

Kuvi AI founders say agents will replace DeFi's app layer, not just assist it

Dylan Dewdney, co-founder and CEO of Kuvi AI, and co-founder Jay Nasr explain why they think intent-based AI agents will reshape DeFi and trading. Recorded in July 2025, the conversation covers their non-custodial agent, its plans for automations and open-source tools, and how they want to onboard users. At the time, they said they expected a product launch and an eventual token launch.

Interview by Ashton Addison Recorded July 2025 6 min read

Key takeaways

  • AI agents can reduce friction in DeFi by allowing users to express financial intents in natural language rather than navigating complex interfaces.
  • Agentic Finance represents a broader concept than DeFi, unifying multiple blockchain tools and applications into a single agent-based interface.
  • The current state of AI agents in crypto is comparable to Ether Delta in 2017, with significant UX improvements expected within 6-12 months.
  • Kuvi AI maintains a non-custodial model where users retain control and must approve transactions, preventing agents from accessing funds directly.
  • Intent-based architecture powered by LLMs can address common DeFi pain points like front-running, slippage confusion, and unsafe contract interactions.

Why agents, and why now

Ashton Addison: What does the merging of AI and blockchain mean for DeFi, and what is Kuvi building?▶ 00:25

Dylan Dewdney: We see two major trends converging. One is people banking themselves and moving toward decentralization. The other is AI agents. Once transformers and GPT-2 could understand intents well enough, we started to see the idea of intent-centric architecture, about two and a half years ago. You say what you want to do and it gets done. Those of us who are blockchain native take the friction for granted. We were fine with all of this friction, but it won't fly when the next billion users come in. This convergence means we can use LLMs to analyze what you want to do in DeFi and then implement it. That's a massive opportunity and real product-market fit. It's not some bloated infrastructure coin or an increase in TPS. It's about people using products more effectively.

Ashton: Where is agentic finance right now?▶ 03:15

Dylan: I'd liken it to the EtherDelta moment. EtherDelta was one of the first DeFi protocols. It worked, but it was really bad to use. It wasn't until Curve and Uniswap came out with really nice UX and UI that DeFi actually worked, seamlessly. I think we're rapidly moving toward what I'd call an agentic finance summer, hopefully in about six to twelve months. Some projects already do the elementary stuff, like spot trades and swaps, by talking to an agent instead of using an application. We think our solution is more efficient. Our big-picture thesis is that agentic interfacing is going to eat up application layers. We applied that to the DeFi application layer, which is especially clunky and full of friction.

We were fine with all of this friction, but it won't fly when the next billion users come in.

One agent instead of a dozen tools

Ashton: How will agents shape DeFi, and how does Kuvi keep users in control?▶ 06:46

Dylan: When we began, there was the term DeFi, and we're merging it with AI. DeFi is maybe 70 to 80% of the pie, but it's not the whole thing, so we wanted something broader: agentic finance. Today you jump between lots of scattered tools, with no unified way to use them. Kuvi brings them into one place and lets you just talk to it. It also helps with the unsafe parts. People get front-run, or copy and paste a contract from the internet and lose their money. Or they hit 50% slippage and have no idea what happened. The agent explains what's going on in a simple way. We're also non-custodial, so you still approve things and stay in control. You can set automations in advance too. For example, when Elon Musk talks positively about Dogecoin, buy some, then take profit at 10% or sell if it drops 5%. You get a confirmation of what will be monitored, and you approve it with your own wallet. That part is still in development. Swapping and asking questions already work.

Ashton: How is Kuvi's approach different from other agent projects, and how far can the technology go?▶ 20:31

Dylan: Some projects, Virtuals for example, give you tools to build your own agent. That's not our approach. Kuvi is one very sophisticated existing agent with a lot of tools. You interact with it and come up with your own prompts and automations. Think of an advanced spaceship that knows how to navigate. You say go to Mars and it gets you there, but you decide where to go. I think AI is, to a good extent, overhyped. LLMs are amazing tools, but they're not thinking tools. I'd bet $10K that there won't be AGI within two or three years. They are great at understanding intentions and connecting tools, though. So we've built many tools behind Kuvi, and the LLM in front only connects the user to the right ones and combines them. Right now we focus on Twitter/X, Reddit is next, and Polymarket is in planning. We also plan to open-source parts of the project, like MCPs, so others can add verifiable tools. It may take a couple of years to reach results better than a full quant company. We're getting there, but we're not there yet.

Who benefits when trading strategies are democratized

Ashton: What is the bigger picture behind agentic finance?▶ 13:09

Dylan: What our angels and our lead seed investor, who took the whole round, caught on to is that this is a new way for people to interact with crypto. It's more efficient, more engaging and gives them more value. The best retail UX in crypto today is a centralized exchange. What we're offering is a zero-to-one difference in the UX and UI of our industry. We're also democratizing access to algorithmic trading strategies. But in the biggest picture, agentic interfacing will eat a significant portion of all application layers in crypto and beyond. For now we're attacking what's possible, which is spot trades and the high-value niche of degens, just as PayPal targeted eBay in the early days. But what we're really eyeing is a sea change in how people interact with value, and they'll do it through agents.

Ashton: Will the best traders be the best prompt engineers?▶ 17:34

Dylan: There's going to be some random lady in Ohio or some random guy in Bangalore who figures out an amazing trading strategy. That might happen in two years, or even less. They'll have access to that power and those tools. Before, it was hedge funds, developers or VC types cobbling together people to find a strategy. Agentic finance democratizes that. It reminds me of the ICO craze, which was about democratizing access to early-stage investments. You couldn't do that before, and now you could. Same here: you couldn't really access algorithmic trading strategies before, and now you can. There's less hype now than in January. But this intersection of blockchain and AI agents went from basically zero nine months ago to billions in value.

Ashton: Some seasoned traders are wary of agents. Are they right to worry?▶ 37:27

Dylan: A lot of people, especially seasoned traders, are afraid of agents. They spent so much time learning these tools, and they don't want a 14-year-old undoing that with a prompt. Whenever new tech comes along, people resist because it makes things too easy, and if it's easy there's not enough money. Here it's the other way around. It will get more difficult, because many more people will be competing with advanced tools, so you need to be even better. In the beginning it's easier, but eventually everyone catches up and you have to think more deeply about how to make money. This game isn't going to end anytime soon.

There's going to be some random lady in Ohio or some random guy in Bangalore who figures out an amazing trading strategy.

Roadmap, speed and community

Ashton: What are the first steps toward that vision for the rest of 2025?▶ 28:20

Dylan: The first step is getting people comfortable with a new behavior. I liken it to going from BlackBerry to iPhone. Today it's obvious the iPhone is the better way, but at the time people said they loved their BlackBerry. Traders say they like being in the trenches and watching the charts. But do you actually like it, or is it something you culturally identify with? So part of our challenge is user behavior. To that end, we run a community prompt engineering program where the public submits example prompts. The total prize pool is about $10,000 in cash. We want people to get used to saying something like: tell me when six of my top ten KOLs talk about the same coin, and then put $500 into it. On the practical side, the foundations are spot trading, swapping and some light automations. Through MCP we can keep adding tools, like sentiment analysis. We're also getting ready for an eventual token launch and growing the community.

Ashton: How do you plan to stay ahead technically?▶ 34:32

Dylan: Dylan: We need to buy some hardware and start building our own models, so we have a moat of efficiency. Right now we're prioritizing speed. Our latency is about a second, and Jay has figured out how to get that kind of performance out of these models in intent analysis like nobody else. Jay: I'm impatient. Other people are working on similar products, and one of the biggest problems they have is that their solution is extremely sluggish. By the time the agent responds, your train of thought has already moved on. You need to be fast and efficient.

This interview has been edited for length and clarity. Recorded July 2025.

Dylan Dewdney
Dylan DewdneyDylan Dewdney is the Co-Founder and CEO of Kuvi.ai, a Toronto-based company building an Agentic Finance Operating System (AFOS) for crypto and digital asset strategies. He has said he has been active in crypto since around 2011, and he took part in the Ethereum ICO.… Full profile →
Share this interviewXLinkedIn
▶ Watching the interview