AI infrastructure needs capital. DeFi has liquidity. We're building the bridge: a new compute-backed yield opportunity, launching ahead of the Autopools v2. More details next week.

Reminder: the snapshot for eligibility for entity participation will happen tomorrow. Remember to migrate your TOKE to AUTO (note: staked AUTO as well as current unclaimed TOKE rewards will be included). Note: TOKE to AUTO migration isn't ending, and AUTO will continue to retain all previous token utility, this is simply for additional new entity participation. Read more below 👇

Yesterday I was asked where I see DeFi "yield aggregators/optimizers" now that LLMs are basically omnipresent. Pertinent question with v2 of Autopools well underway, so a couple of thoughts: People think of "let an agent handle it" way too broadly, especially in DeFi. For many of the required tasks, LLMs aren't the optimal tool. That goes double when you're dealing with a wide universe of opportunities across different markets, position types and protocol mechanics, where the headline numbers usually aren't directly comparable in the first place. Allocation, execution, rebalancing are closed-end problems. There is a right answer, and a deterministic solver gets there better than a language model. "Not an LLM" doesn't mean "not AI". Jev is the obvious counterexample. Typed decisions, no text generation. It won't solve an allocation, but it's a plausible fit for the screening layer that feeds one. The LLM earns its place somewhere else. It's a conversational interface that walks a broad opportunity set with you and distills it into an executable strategy, with destinations, constraints and trade-offs made explicit. It helps you work out what you want against what's actually available. Once the strategy is settled, the solver takes over. Allocation, rebalancing, margin maintenance, constraint enforcement. The LLM translates intent, the solver executes and maintains it. Two different jobs. So the yield aggregator of tomorrow can't just be a vault with customizable strategies. It has to be a data provider and solver too, exposed through an interface that serves both sides. A person working through the UI or an LLM, and an autonomous agent calling it directly. Either way the job is the same, turning stated intent into a running strategy. Which ultimately makes it infrastructure for the agentic web.

Got something bigger like perps, funding-rate arb, a full onchain ETF? Leave a comment in the replies.

👇

Props to our friends at @Balancer for working hard to keep Balancer v3 safe from the next generation of threats 👇

DeFi security requires constant diligence, and trust is earned through continuous verification. We're grateful to @hexens for completing an audit of our latest security hardening updates, ensuring the protocol remains robust and safe as it continues to evolve. Read more below 👇