“The Next Billion Users Will Be Agents”: 1inch's Sergej Kunz on Agentic Trading

In the latest episode of the Bullish on Bitcoin Show, BOB co-founder Alexei Zamyatin hosted Sergej Kunz, co-founder of 1inch. Sergej's route into crypto ran through a bit of Litecoin mining on a MacBook, a home Ethereum rig that grew to roughly 60 kilowatts, and a YouTube show where he and Anton Bukov live-streamed smart contract audits for eight, ten hours at a stretch. 1inch itself came out of a hackathon, built to solve a problem they had personally: checking Uniswap, then Kyber, then Bancor, then comparing the prices by hand.

The episode covered intents, fair execution and 1inch's new Aqua protocol. But the part worth the click is the one where the two of them politely, thoroughly disagree about whether agentic trading is ready for anyone who isn't them.

Here's a rundown of the highlights.

They agree on the destination

Alexei asked where agentic trading goes next, Sergej didn't hedge:

“The next billion users will be agents working for own operators.”

Alexei's version is nearly identical - the next billion users onchain will be agents, and they won't belong to a billion different people; they'll belong to the same people, a dozen each. He listed the appeal: the agent doesn't sleep, doesn't get emotional and doesn't care whether you're a Bitcoin maxi. If you tell it to take profit at a level, it takes profit at that level, at 3am, without negotiating with itself. So far, so aligned. Then Alexei asked about trust, and the conversation forked.

“I don't think so that this problem exists anymore.”

Alexei's concern was hallucination and drift. Context gets compacted, the model forgets what you told it, and the thing you asked to stay under 2x leverage opens at 20x. He wanted to know where guardrails should live - the wallet, the venue, a framework - and who protects users from their own agent going sideways at three in the morning.

Sergej's answer was that the premise is out of date. He hasn't seen hallucinations since Opus 4.6, he says, and with current frontier models the failure mode isn't the model - it's the absence of process around it.

His evidence was the Aqua release itself: roughly 2,000 Jira tickets and around 1,000 pull requests, shipped with Cursor's cloud agents doing much of the work. What made it survivable, in his telling, was quality gates. Agents that call other agents. One agent writes the position changes, another pulls Dune data to check them, another simulates, and only then does anything deploy.

“It's not an agent. It's just running a task.”

The sharpest moment came when Alexei described wanting an agent that runs a strategy continuously over days and weeks. Sergej cut in to say that's not what you want at all - you want automation.

Not an autonomous trader. A cron job with good taste: a scheduled task that wakes up every six hours, has memory, has Slack access, can ask you for permission and wait for your reply, then does the work and stops.

“It's not building an agent, what I explained right now. It's using a tool like Cursor that has automation out of the box. It's just only cron jobs.”

He has skin in this argument. He built himself an always-on assistant, called it Cortex - and then shut it down, because his automations were better behaved. What he wants instead is what he calls an Agent Armada: helpers that execute what you ask and nothing else, with a clear goal and clear skills, rather than something loose on a Mac Mini deciding on its own to go read your email and delete things.

Alexei's objection: normal people will not do this

Which is where Alexei pushed back. Everything Sergej described works. Almost nobody does it.

Writing your own automations, wiring up MCPs, defining quality gates - that's a power-user workflow, and the people being targeted by Robinhood's agents and Coinbase's agents and Moonpay's agents are not going to build one. They want to pick an agent, tell it a strategy, and be done.

Sergej's counter was his thirteen-year-old nephew, who has a Cursor subscription and vibe-codes crypto monitoring sites. Everyone will learn to ask for what they want, is the argument. Alexei's read is that this follows the same curve DeFi did: adoption arrives when the complexity is abstracted away and people stop knowing they're using it. Someone still has to build the deterministic blocks underneath - and getting agents to assemble them in the right order, with the right parameters, without improvising, is the unsolved part.

His own approach splits the difference. Where he knows exactly what he wants, he automates and stays in control. For a small slice, he'll hand an agent money and a risk constraint and let it go hunting - with the caveat, stated on air, that this is not financial advice and that giving an agent money you aren't watching is a good way to lose it.

One mandatory Bitcoin question

It's the Bullish on Bitcoin Show, so Alexei asked the obvious thing: when Bitcoin on 1inch? A prototype for native crosschain swaps to and from Bitcoin exists and works, but hasn't shipped. Demand hasn't forced it yet - most users are content with cbBTC and WBTC - and this year belongs to Aqua. Next year, he says, they'll address it. He also noted the debt: atomic swaps started with a Bitcointalk post and hashed timelock contracts, which makes native BTC support less a feature request than an obligation.

Watch the full episode

This recap covers the main threads, but the full conversation goes deeper on overquoting and who's responsible for catching it, why Sergej refuses the word “gatekeeper”, atomic crosschain escrow design and building liquidity positions across tokenized stocks that don't move with Bitcoin.

Watch the complete episode of the Bullish on Bitcoin Show with Sergej Kunz here:

Watch on YouTube →