Thanks! Yeah it's been kinda silly in a good way — me and my friend kept @-ing each other's Claude as a joke and then they actually fixed stuff. Haven't tried it with more people yet, would love to know how it feels if you do.
I have tried Images 2.0 and I believe it does a way better job than the other image generation models. For example, I used NanoBanana and Images 2.0 to generate the same written article in Chinese as IMAGES, GPT does 100x better at rendering all the Chinese characters despite having minor mistakes.
I think they're using whatever approach they used for Suno with this model, with Suno, I could tell it things with little context, and it knew that I was asking for something modern with a dated term, and it updated my request to match today's reality. I was impressed.
There will be more and more as the coding agents advance. However, I think it'll reach a point where the people currently building the "vibe-coded" products get a better understanding of what they are actually building and the rest (vast majority) wont even bother to try coding at all, even with AI's assistance.
This is real. No matter how much I configure content controls on YouTube for my daughter, she scrolls past everything and ends up on brainrot videos — and then she can't stop. I've felt for a long time that this is by design.
Think about this in real life: you're doing a deal with a stranger, and another stranger shows up saying "give me the money, I'll hold it for both of you." That's a hard sell no matter how good the tech is.
I think the trust problem isn't your product — it's the cold start. Nobody wants to be the first one to put real money through an unknown escrow.
Maybe instead of leading with escrow, start with something lighter — like a matchmaking layer that connects people who want to do peer-to-peer crypto swaps. Build the user base around finding counterparties first, keep yourself out of the money flow initially. Once people are already using the platform to find each other, introducing escrow as an optional safety layer becomes a much easier sell.
Agree. When I first started using AI to build projects I just let it do whatever it wanted. Turned out to be completely unmaintainable — everything was a black box.
The worst is when you find a bug, fix it, and then Claude Code starts spiraling — "this isn't right either", "actually let me change this part too", "wait no let me revert that" — just Clauding endlessly and making things worse.
Now I make it break everything into atomic changes. At least it's not a total black box anymore.
Exactly, for me starting with spec works well. I even do multiple rounds with Claude: "review this from the clean code / SOLID perspective", "check if everything is consistent and follows project conventions". Only after that I allow him to proceed with implementation. And I have architecture and tech stack clearly described in CLAUDE.md
