Wrong anchor. Manuel is a legend in smart contract security, so when he issues a PSA to exit DeFi because AI agents make it fundamentally unsafe, people listen. But his conclusion anchors on the wrong premise. Credit card fraud was once considered an existential threat to digital commerce. The industry didn’t solve it by eliminating fraud - that was never possible. They won by engineering the per-incident blast radius down to what the system could absorb: $0 cardholder liability, tokenization, real-time detection, and issuer insurance. Trillions flow through card networks daily on top of a permanent attack surface because they anchored on containment, not elimination. History repeats this pattern across aviation, the early internet, and traditional banking. Each was once labeled doomed or fundamentally unsafe. Each won, not by becoming error-free, but by iterating until the system could absorb the failures it couldn't prevent. Yes, AI has lowered the cost of attack. That's real. But as many have noted, AI is just as available to defenders. Auditors, security researchers, whitehats, and protocol teams are using the same tools. And as @StaniKulechov pointed out, DeFi has made generational progress on the defense side: better risk engines and lending market structures, formal verification, oracle improvements, cap management, circuit breakers, automated monitoring, SOC2-grade opsec. None of this existed at scale a few cycles ago. The ecosystem has been building the right infrastructure to guard against attacks, similar to what credit card industry has gone through. AI is not our enemy; hackers are. The anchor isn't "bug-free". It's "no single bug drains the protocol." If you design every system with the assumption that a vulnerability will exist, you stop trying to build an unbreachable wall and start engineering the blast radius down to what the network can safely absorb. The real PSA isn't to exit DeFi. It's to build it right. DeFi will win.