launch day

I cut my monthly AI spend 10x from $2,000 per month to $200 by using Codex instead of Cursor For a while I couldn't get behind the chat window UI vs. a real IDE. But, now Sol is so good and openAI gives you SO much usage for $200 that at this point if you're not using Codex you're trolling My workflow involves a single 'boss' agent who manages ~5 other chats at a time, and it works seamlessly - agents talk to each other, I talk to the boss and it reports to me, shitload of work gets done quickly and at high quality Can't imagine how good it will be once we have Astra

After going through the hiring process for an engineering role on the employer side, I am 100% convinced that the “software eng job market sucks” phenomenon is mostly the fault of the applicants, not “AI” I’ve reviewed roughly 100 applications over the last 2 weeks: 50% were thrown out because they didn’t respond to our AI-submission filter question. They either used an AI tool to apply or didn’t read the listing. Another 30% gone because they have 0 projects outside of schoolwork, empty Github (of those with a short work history, not counting experienced devs working in private repos) Another 15% gone because their personal website is a self-report that they don’t understand design or performance at all (or both). We’re hiring for a frontend heavy position. Like a restaurant, if the front of house is dirty - your mind starts to wonder what it’s like in the kitchen. Just three conditions, and already 19 of 20 applications have been thrown out 95% failed three extremely controllable, basic filters before we even got to seriously evaluating engineering ability If you’ve been looking for dev work for months without success, before concluding that the market is hopeless, take a hard look at what you’re actually putting in front of employers. Read the listing carefully and show that you want THIS job not A job. Take the extra time to work on a genuinely GOOD personal site. Even if your work history is lacking, if your site is good that’s proof enough that you are capable of writing decent frontend. AI makes this 10x easier and so you have even less of an excuse not to. And yes - some genuinely good candidates are out there just having a bad beat. Karen from HR with a biology degree throws out your application. Some LLM filter screens you out for an indiscernible reason. I totally get that and empathize sincerely. The total number of open listings is shrinking while the pool of applicants is growing. It’s hard. I had it way easier 5 years ago. BUT - you need to recognize the situation you’re in. If there’s 100 jobs out there and 1,000 applicants - you NEED to be in the top 100. Ideally the top 20. The market is harder. You cannot control that. You can control whether you are the person who stands out in a harder market. The internet gives you unlimited free education. AI gives you an on-demand tutor, pair programmer, designer, and reviewer. None of this guarantees you a job, but there has never been less excuse to submit an application that gives an employer no reason to choose you. Stop optimizing for number of applications submitted. Optimize for being impossible to throw out. You do not need to beat the job market. You need to beat the other applicants.

emails you love to see on a wednesday afternoon after choosing Resend:

cursor employees are using fable 1M context to adjust padding values you’re not ripping enough tokens

can we use 1M context window yet? I tried circa ~opus 4.5 but the model got cte past like ~300K, can the new models handle it

The American experiment succeeded because the U.S. has an unusually high concentration of people at both extremes of the bell curve Lots of crazy people, lots of geniuses

if I worked at Cursor this is the first thing I'd implement

it’s the small things

imagine reading this in 2024

how would you improve this waterfall chart's design? anything that could be clearer?

is Claude Tag not just River for everyone? not seeing River brought up at all, Shopify ahead of the curve

would you believe me if I said this component took $22 worth of opus tokens to build correctly

me: yeah so whenever we’re planning features now we just ask jarvis to write the first draft then we come in and refine it through a q&a back and forth until we agree on a final plan my mom: who’s jarvis?

request for brand that will print money: - premium oral care - hero product is a beautifully designed water flosser - apple-esque industrial design and branding - proprietary cleaning capsule you drop into the reservoir every use (think non-foaming mouthwash designed specifically for water flossers) - cleaning capsules sold on subscription $29/mo, stupid margins - expand into toothpaste, floss, toothbrushes, whitening, etc - sell the whole routine on subscription $59/mo - profit