bunker mode
the guy below is an ETH researcher for 9 years and is followed by my goat @VitalikButerin
eth page retweeted it aswell.
this just might be the best narrative on mainnet in such a long time (yes am a fan)
0xf4f986ba1362ed61f9aeb683be20f6219056d2c1
nerd retweeted
We’ve been building heads down behind the scenes.
One of the biggest updates arriving this week is Phase 5: Progression.
⚠️ Class Advancement
Reach Level 10 to choose your path.
Current classes:
⚔️ [Kael] Swordsman → 2nd Class
🏹 [Veira] Hunter → 2nd Class
🛡️ [Brutus] Champion → 2nd Class
This is the first step toward preparing everyone for the upcoming dungeon and future endgame content. Stronger enemies will demand stronger builds, and Class Advancement is where that journey truly begins.
We're only 2 weeks since launch, and we're already rolling into Phase 5. Rather than making everyone wait months for meaningful updates, we'll continue shipping new content and progression systems as they're ready.
biggest weakness i see rn is that agent's learning is still mostly manual and operator guided.
the logs, memory and price history are there but it still doesn't have real self improving weights or string pattern learning yet.
i still might call out a few teams who keep rugging on the EVM chains after this transition, but for now..
i’m in full ‘just build stuff with @claudeai’ mode.
when i talk about building security at ground zero.
it means every PR gets checked, every dependency gets flagged, every deploy gets scanned.
security isn't a gate at the end. it's the foundation you build on.
i have looked into so many recent hacks and the post mortem is always the same.
'we didn't expect this attack vector.' dude you didn't expect it because you never looked. 😂
been doing security and the pattern is always the same.
team moves fast, skips scanning, gets audit last minute, auditor finds 30 things that should've been caught in the pipeline
security should not be an after thought, build security at ground zero
everyone's obsessed with prompt injection.
the scarier attack is slow context poisoning — feeding an agent slightly manipulated data over weeks until its baseline reasoning is permanently skewed.
no single request looks malicious.
attackers are using ai to generate polymorphic exploits that mutate on every attempt.
defenders are using ai to... highlight unused imports.
we're cooked.





