Understanding the design principles behind, @AmpCode, one of my fav AI products, is truly enlightening. Thanks @jxnlco @beyang @Sourcegraph

Flashbots is significantly ahead of its time in multiple dimensions.

I built a Raycast extension that allows you to create, list, edit, and delete Cloudflare email aliases. I use @mmvlad Chrome extension daily but this is much more user friendly. Repo: Store:

Cloudflare outages are sometimes the exception you welcome. - Take a break from work - Read a thorough post-mortem on the incident.

It has almost been a year since I left the crypto space in my professional work. Here's what I have done so far: - have experience working on a 1M+ LoC enterprise codebase (partially, not fully). - have a solid understanding of 'product'. previously, I had no idea about product development. only knew narratives and VC appetite. - wrote a very serious paper about my current thesis on crypto, which wouldn't have been possible a year ago with my previous mindset. - fully configured an AI productivity + dev workflow. by the time I saw crypto people start using AI tools in my feed, I had already mastered them. - have a decent understanding of infrastructure (cloudflare, AWS, vercel). - constantly blogging. I don't think it would have been possible if I were still working in crypto. If you're also questioning your involvement, changing direction might be a good idea. You most likely didn't miss anything; you'll be ahead.

Most tokenomics presentations: - 30 slides about supply emissions - 0 slides about demand generation mechanisms This is backwards. We need to flip the entire conversation. Supply is easy to control. Demand is what separates winners from losers.

If you’re short on L2s, you should build an unified escape hatch frontend for all L2s than bridge to your own L1 (Monad, Solana etc.)

increasingly difficult to recommend crypto to young talent relative upside is down 10x relative to AI and too much of crypto business revolves around scheming rather than learning how to create value

My hot take: PMs are the bottleneck in AI-first products b/c many engineers find "looking at the data" and doing evals "beneath" them PMs are more willing to do qualitative analysis on LLM traces, build theories of bad LLM behavior, aka evals---and evals are the bottleneck

reminder