Vlad went on The Iced Coffee Hour for almost two hours today. Most of it is company stuff you can skip. Here's takeaways that we thoughts were most important and interesting.

Agents refuse to trade sometimes, and the reason isn't what you would assume.

This was the biggest standout. People hook an agent up to a brokerage, tell it to place an order, and it comes back saying it doesn't think that's a good idea. Everyone assumes it's some built-in safety layer.

Tenev said it usually isn't. The model isn't saving you from a bad trade, it just doesn't do it. His explanation: training data is full of agentic coding traces and has basically none for trading, because nobody's done it at scale before. So the model has nothing to pattern-match against and hedges instead.

Which means it's a gap that closes as more of this activity happens, not a wall.

The agent thing is already bigger than most people think.

They expose an MCP server so your agent gets the same tools you'd have in the app. Equities first, options live, crypto announced. Over 100,000 agentic accounts already, doing real volume, with people wiring in outside data sources to build strategies. He said he doesn't really look at their returns, which is either refreshing or slightly alarming.

The framing he used that I thought was right: before LLMs, running any kind of algo strategy meant being a programmer, building integrations with legacy financial plumbing, and paying about $10 a trade for market access. That's the barrier that has seemed to have disappeared now.

Memecoins and tokenized stocks are getting composed together, and not by the company.

We made a post about this the other day so it interested us to see him acknowledge this as well.

Robinhood launched a chain with stock tokens on it tokenizing TSLA, NVDA and others, live in 120+ countries, US excluded. Once those things are on-chain they act like any other token: send them peer to peer, LP them into Uniswap pools, post them as collateral, borrow against them.

The interesting bit is what developers did with it. He said people built protocols where holding a memecoin airdrops you stock tokens, and connected memecoins, core crypto and equities into products the company never would have designed itself.

Basically memecoins now have utility and distribution rewards via tokenized stocks that they are being paired with vs the standard USDC/ETH/SOL etc..

Bitcoin is "the original memecoin."

His argument was that the brand is the moat and you can clone the tech endlessly, but nobody can take being first and most trusted. He's long-term bullish on BTC of course.

His take on security atm.

Pre-AI you could sort of rely on vulnerabilities being expensive and annoying to find. Now spinning up another job is cheap, so nothing is safely obscure anymore. They red-team themselves with AI tools pointed at their own infrastructure. Relevant to anyone self-custodying or running anything with keys in it.

Short version: agents can trade now, and when they refuse it's a training-data gap rather than a safety feature. Tokenized stocks are live in 120+ countries and devs are already composing them with memecoins in ways nobody planned. Bitcoin is "the original memecoin" and he's long-term bullish. His bubble signal is everyone suddenly wanting the same thing at once, currently going off on Korean stocks and the chips/energy trade.

Where we think this goes in regard to AI x Crypto

Our personal view: It's clear AI and agentic trading will have a massive impact in the space. We've seem tons of consumer apps hitting the markets that make trading memecoins and crypto more accessible. However, we believe the breakout app this cycle is at the intersection of crypto and AI, and it looks a lot like the ChatGPT of crypto where trading collapses down to a prompt.

Crypto is where that lands first and it isn't close. Markets are 24/7, the rails are open, and an agent holds keys instead of asking a broker for permission. Every constraint that makes agentic equity trading awkward is either absent or optional on-chain. We fully back this belief, which is why we have been building exactly this for the past year and we have seen a couple of other apps working towards the same. AI x Crypto consumer applications will be one of the largest waves the space has seen, and we're still very early.

Would love to hear thoughts on the interview! It was quite a long one.