Steve Butterworth

Relentlessly Simplify ™

@doodlestein @yongfook 🤣

@stevewillbe @yongfook I don’t know, seems like a dangerous game to play now, especially if you have a corporate W2 job.

I don't think AI will be the 10x or 100x multiplier that people say, because I think there are natural checks and balances in place regulating everything. You can code 10x faster now? Cool, do that for a week straight and you'll fry your brain. You will need to take breaks and regulate yourself. You can build a product 10x faster now? Cool, but it takes just as long as it always did to get people to care, get customers, etc. You can tear through the entire backlog in one day? Cool now there's no more work to do and everyone is fired. Maybe lets not tell the boss this is possible. You can generate any image of anything you want? Cool, but your friends can only take so many custom meme images before they disown you. You can send 10,000x more AI-enhanced "cold outreach" emails looking for clients? Sure but it takes just as long as always to gather requirements, discuss timelines, etc. None of that is any faster. Only a very tiny % of giga-chad geniuses will have both the intellect and the agency to employ AI in a way that actually 10xs or 100xs the value that they generate. The rest of us will be limited by organic or bureaucratic guard rails. Personally I'm very grateful for the role AI plays in my life. It allows me to code 10x faster with the time I allocate, which frees up other time to do things I enjoy in my personal life. My output is more than before, not 10x, maybe 2x. And my quality of life has improved. That's well worth $200 a month!

@doodlestein @yongfook Yep it is. So many incredibly varied ways we can succesfully navigating this thing we call life

@stevewillbe @yongfook This is the same argument as “Why should I work hard when I can slack off and skate by with the minimal amount of effort?” And yeah, if that argument resonates with you, good luck with that.

@doodlestein @yongfook Liberating to mange 52 when we can manage just 1 and have more free time than ever to enjoy its rewards?

@yongfook I don’t sell anything actually, it’s 100% free and open source. Appreciate the snark though.

@sobedominik Like others mention /rc is good when it works but cloud layer does seem flaky at times. Feels pretty beta. Tailscale + Moshi is my fallback.

is there any great tool that lets me easily keep track of my Claude Code instances so I can review, interact and approve proposals while I sit on the toilet from my iPhone? 📲 👀

@milalolli @charlierward 🇬🇧 changeflow.com 🚀 London

If you are London based and Londonmaxxing building stuff. Drop a 🇬🇧 below 👇 I’m forming a group for IRL events and I want to see you there

@tadasgedgaudas @iannuttall I’m a bit more optimistic about pricing than this. The competitive landscape shows the Chinese labs catching up and the Chinese fabs not far behind. The cut throat battle among US labs isn’t stopping soon either. All good for the consumer 🤞

@iannuttall on it!

I am very happy to be the product if I get $5k+ in compute for $200 just remember: you gotta make hay while the sun shines...

@yongfook Same strategy. Let others ride the hype cycle and fix the bugs and then I’ll play with what comes out the other side.

You can set your watch by the fact that if I get into some "new tech" that probably means it's about a year old already. I got into Claude in Jan this year. And just installed Ghostty and I'm like "heh, this is cool".

@forgebitz @ImSh4yy @OfficialLoganK @GoogleDeepMind I’d be happy with a Gemini 3.1 Flash Liter but I can’t see it happening

@forgebitz @ImSh4yy @OfficialLoganK @GoogleDeepMind Crazy decision. For automated high scale use I need a very cheap half decent model. Flash 2.0 and Flash 2.5 Lite were these. They’ve taken themselves out of this market so we’re left with the lower end OSS Chinese models there now and grok 4.1 fast.

Introducing Gemini 3.1 Flash-Lite 🔦, a huge step forward on the boundary of intelligence, beating 2.5 Flash on many tasks.

@OfficialLoganK @GoogleDeepMind With those price differences you’ve just axed the Flash Lite USP. This was a reliable, cheap model for scale, no more.

@ArtemXTech This is remarkably similar to my approach. QMD everything + recall for searching old sessions. Works very well

@yongfook Love how it’s transformed how I work - Super scared on how transforms the world of work and therefore modern civilization 🤷

Thanks to AI my mood swings multiple times a day from the sheer exhilaration of pure build mode, to deep existential dread. Exciting times.

@helloitsolly This is good 👍

Final #buildinpublic thoughts on hiring an engineer Engineer will need to document as they go, like everyone else (including me). Updates might include: -> Reproduced the bug locally -> Found the issue: image resizer not handling SVGs correctly -> Tried a CSS fix first before touching the component -> CSS fix didn't work, going deeper into the component -> Fix working locally, pushing to staging -> PR open. Tested on Safari, Chrome and Firefox (these could be automated -> @linear ) Automate Gitbhub to @linear issue status updates There will be a daily check in on Slack I am going to remove the Svelte requirement I am going to keep pay at $9,000 a month and open up the option for a profit share Continue to share the SOP in the job description I will keep no recurring meetings but have a one-week onboarding with as many 1:1 calls as required, plus ongoing calls as required I will try hiring locally in Cape Town and collabing in person I will do a deeper review after a week and a few weeks No passive Slack messages - let the person do their work or not, let go of control

@helloitsolly @IsaacHinman @adamcollins I don’t think this is mid in the UK. I’m a senior contractor, would charge double and would want more than 2 days to familiarize myself with a codebase and roadmap before diving into a single ticket.

@IsaacHinman @adamcollins Mid in Europe, UK and Canada where we are hiring

Another engineer quit midway through a trial $9,000 a month and apparently keeping Linear updated as you go is too much Listed twice in job description Designed to facilitate remote work and fewer meetings

@helloitsolly @dozibe That’s such a developer request, you dodged a bullet, that spectrum of developer is prevalent and definitely not what you need. Communication is everything

@dozibe He was highly skilled but wanted deep autonomy from day 1 from what he's said

@simonw Let there be Lite!

@simonw And we haven’t had a Lite model since 2.5

I didn't think we would get Gemini 3.1 Flash Image before we got Gemini 3.1 Flash!

@yongfook And it’s all Rails too. One of our few public sector technology success stories but it’s one to be proud of

I just signed up for the UK gov "one login" and it's an absolute masterclass in UX. No sarcasm. It gracefully handles handoff to a mobile app for facial verification, identity document capture (with multiple options) and handles all sorts of edge cases. Every time I felt "I bet if I click this the form will reset and I lose my data" it all just worked. Super clear language and UX, at no point was I frustrated or lost. The UK might be in decline as of late, but we still have greatness within us somewhere!

@yongfook Agree. The best models to chat to don’t need any of that. If you’re trying to build reusable prompts for your app and keep costs down a speed high for repetitive inference by using cheaper models you need alot of clever prompts engineering.

I’ve never once thought “oh there must be a special way to ask this”. Maybe this is due to me using Rails though. Everything is in a predictable place, I don’t have to explain where things live, it just knows.

Hot take: prompt engineering is BS I just talk to Claude as if it's a human and it works great. The idea that among 1 trillion parameters, Joe Nobody has figured out the optimal input X to get outcome Y is laughable. Just talk to the damn thing bro.

@iannuttall @Shpigford Tried handy, it’s basically that

@Shpigford I mostly use monologue but was gonna fork this one and make my own using parakeet github.com/watzon/pindrop

this feels like a space that's changing obscenely rapidly but, what's the best/fastest/most-accurate voice-to-text app for macos?