Seif

Ended up publishing the Claude Code proxy. Vibe coded, but still a really fun tool to use to understand how this stuff works. You can see things like system prompts, implicit system-instruction injections, and how it proactively calls the Read tool when files change mid convo.

Operational excellence is under appreciated in general

@JonasScholz19 Have you tried Claude Code’s review plugin? Pretty solid. I’ve also recently been using Pi and writing my own Skills for code review and it’s been great.

is there a good code review tool that i can just use with my existing codex/claude subscription?

@sebo1738 @Shopify @shop Lmao congrats dude. 2 lessons here - believe in yourself; spurs are shit :)

Spurs might be dogshit but I called it over a year ago 🙏 Am now a product data scientist @shopify & @shop shoutout @seifAndSoundTO

Hey @bcherny @trq212, why is this block repeated twice in Claude Code’s system prompt?

@fahdananta Legend, all the best dude 🙏

I want to share the story behind why I joined Opendoor (It might be a little long. I'm not media trained, so this will be straight from the dome and I'm just going to hit send at the end) The last 4 years I was basically retired. I started a small fund and was having a great time investing. I live a pretty modest life so I never really planned to work again. Over the years so many companies reached out. I met with the execs of large public companies, celebrities and athletes that were starting software companies, and so on. Some I really considered, but always ended up say no. It wasn't about the money, they offered insane amounts but I just liked my simple life. I'd drop my kid at school, bike to the studio space I share with a bunch of my friends, spend the day reading, investing, coding, grab a coffee, and head home. I've always been fascinated by asymmetric risks. Usually I do it from the sidelines like any other retail investor. I've also seen how often consensus is flat out wrong. I was at Shopify as a lowly PM when this short seller report came out in 2017 calling the business a pyramid scheme. It was crazy to reconcile what I was seeing in front my eyes, a company operating at maximum excellence, hypergrowth, etc and everyone else repeating whatever they heard. Turns out they were wrong. I saw this a few times over. Before Shopify, while I was building my first startup, everyone thought we were idiots. These were credible people with serious sounding job titles at hot tech companies. We ended up getting acquired by HubSpot. Turns out they were wrong. When Carvana got decimated, and even today, people love the narrative that it's a fraud. Turns out it's a great business that people just don't understand. I was deeply invested in the business and knew the business was turning around while at the same time the media was publishing fear. Turns out they were wrong. After Robinhood got decimated, same thing, people were unable to consider that the business can change and evolve. I posted an experiment and my P&L on twitter and people DMed me calling me an idiot, over indexed, going to 0, and so on. Not just the anon accounts but real people I know and respect. Turns out they were all wrong. I guess I'm not just fascinated by asymmetric risk but I'm addicted to situations where I think everyone else is wrong. At Opendoor, it's incredible man, I get to actually do the work, not just watch from the sidelines. I get to work with people that are at the top of their craft and fully committed to the mission. So, when the opportunity came up. It was an instant decision. I "un-retired" and am all-in on the biggest bet of my life. We'll see how it works out but you already know where I placed my chips. At the end of the day, none of this shit matters. We just have to ball out, have fun, and pursue excellence. I'm not doing this for the money, I already have more than I can spend in my lifetime. I'm doing this for the love of the game and most importantly, I'm doing this because I think everyone else is wrong.

@JonasScholz19 Need a few more of them to fix that state drift

had my second Celsius and the terraform started to talk back

@JonasScholz19 lol 100% agree. What were you trying to do tho?

never use llms for low level networking stuff lol

Umm I don’t get the gpt 5 hate… it does pretty well at helping you dig and reason through a problem.

lol I was wondering why the proxy was getting so much traffic when it first went public

Such a good read.

So bad it had to be intentional

Also shoutout @freeatnet. Feels good not seeing a random issue pop up for this sorta thing

A first contribution to the proxy 🥹

So you can route Claude Code subagents requests to different model providers. Routing works really well. At least until the next Anthropic update lol

Getting Claude code subagents to work with different models has been both annoying af and pretty fun. Going to release it on the proxy next week.

@preshdkumar The best kind of channel. Love to see it man

whats your main growth channel? "the sauna"

@evansammccann @turbopuffer Sweet, will check it out for sure

@seifAndSoundTO @turbopuffer recording a pod episode with turbopuffer next week

Folks in Canada need to stop sleeping on @turbopuffer. The most exciting Canadian tech startup out there.

@sergical Congrats!

you know that feeling where you can call your parents and tell them not to worry about their mortgage for a while? Highly recommended I’ve been chasing it for so long, it’s been the main motivation for everything I do. So grateful for all the love, really means the world to me. Will keep building 🫡

@nicoritschel It’s pretty much still a Task tool call, but with the added feature of you being able to supply a custom system prompt for each subagent as well as restrict tool usage.

Oh I think I overlooked that each role maintains its own context and can be re-summoned; okay that’s neat

Claude Code is great but the new custom subagents that have personas feel a little too characterai for me

The male urge to build an agent orchestrator