Justus
709 posts
Justus
@ibuildproduct
AI/Bootstrapped/profitable/growing
London/NY/SF เข้าร่วม Temmuz 2017
57 กำลังติดตาม175 ผู้ติดตาม
Hey builders:
Does your product solve your own problem?
English
So after my last tweet icon.com now became the founder's Tinder profile
@levelsio@levelsio
Icon, the AI Admaker, just went bankrupt They paid $12M for the domain Icon.com and now it's dead
English
He's right. The Claude macOS app is the most broken piece of software I've used in my life
Tibo@thsottiaux
@Yuchenj_UW Not yet, and that’s why it works extremely well
English
English
@ibuildproduct @levelsio @a_klofat @itsolelehmann Nice, how would you tackle visa should you like to stay in the US?
English
impossible to find anything decent to live in Munich for short term rent (4-6 weeks)
absolute dogshit apartments going for 5-8k/month wtf
English
@_cstoeckl @levelsio @a_klofat @itsolelehmann I‘m still undecided. I’ve spent 5 months in NY, SF, and London over the past year to test them out.
Now I’m gonna go to SF for 3 months to get a better feel for it, I was only there for a few weeks last year.
English
English
GIF
English
@a_klofat @_cstoeckl @itsolelehmann My German entrepreneur friends are all leaving Germany, it's not a place to build a business or live a good life anymore, sadly
English
@levelsio @latentsauce Isn’t it in the base context? I.e. Claude.md or agents.md in the folder it is launched from
English
Justus รีทวีตแล้ว
BREAKING: Your calendar now has an assistant!
There is a better alternative to calendar links now!
SkipUp.ai is an ai scheduling assistant that handles complex scheduling for you. Simply CC skip@skipup.co and it runs the conversation for you. Works 24/7 over email and has a 1-minute average response time.
It can also:
- Follow up, confirm, reschedule, and book meetings
- Learn your preferences (buffers, windows, no-go days, etc.)
- Handle time zones
- Multi person scheduling
- Multi-calendar integration
- Connect to your CRM or agent workflows
Give it a shot with our free trial.
How it works🧵
English
In my experience, data analysis and comprehension is the hardest area to use reliably. I’ve been using it explicitly for this purpose and consistently see inaccuracies: it takes shortcuts, makes assumptions, and degrades quickly once the data task has any complexity.
I’ve tried verifiable planning steps, smaller scoped tasks, explicit testing, and incremental execution, yet I still have to constantly audit outputs and correct data manipulation errors. I see the same pattern with data transformations and migrations.
One practical example: I was reconciling two large knowledge bases (mostly JSON and Markdown) serving 3k+ engineers, to establish a single source of truth for standards across both Git projects. This required semantic comparison, detecting divergence, and preserving intent across many incremental steps. In this scenario, the model repeatedly inferred structure, made unstated assumptions, and degraded across steps unless every transformation was manually verified.
I’d appreciate more detail on the guardrails you use in practice to prevent this deterioration in data-heavy or analysis-driven workflows, particularly how you constrain assumptions and ensure fidelity across multi-step transformations or assessments. I’ve also observed that some other models appear stronger in this area and would be interested to know whether and how you benchmark against them.
Thanks 🙏
English
I'm Boris and I created Claude Code. I wanted to quickly share a few tips for using Claude Code, sourced directly from the Claude Code team. The way the team uses Claude is different than how I use it. Remember: there is no one right way to use Claude Code -- everyones' setup is different. You should experiment to see what works for you!
English
@nummanali They are not fixing fast. It has been buggy for a while.
Also you can ship fast without it being this buggy, especially when you have billions and supposedly the best engineers in the world.
English
@ibuildproduct Can totally understand the sentiment
But I believe shipping fast and fixing fast will be the norm
And as models improve, there will be less issues
English
Creator of Claude Codes reply to Andrej Karpathys’ thoughts on coding with AI
What I take from here:
- Opus 4.5 is writing 100% of code at Anthropic
- Because they dogfood, they know how to improve the model
- Many still haven’t switched to this method of coding but will soon
Boris Cherny@bcherny
As always, a very thoughtful and well reasoned take. I read till the end. I think the Claude Code team itself might be an indicator of where things are headed. We have directional answers for some (not all) of the prompts: 1. We hire mostly generalists. We have a mix of senior engineers and less senior since not all of the things people learned in the past translate to coding with LLMs. As you said, the model can fill in the details. 10x engineers definitely exist, and they often span across multiple areas — product and design, product and business, product and infra (@jarredsumner is a great example of the latter. Yes, he’s blushing). 2. Pretty much 100% of our code is written by Claude Code + Opus 4.5. For me personally it has been 100% for two+ months now, I don’t even make small edits by hand. I shipped 22 PRs yesterday and 27 the day before, each one 100% written by Claude. Some were written from a CLI, some from the iOS app; others on the team code largely with the Claude Code app Slack or with the Desktop app. I think most of the industry will see similar stats in the coming months — it will take more time for some vs others. We will then start seeing similar stats for non-coding computer work also. 3. The code quality problems you listed are real: the model over-complicates things, it leaves dead code around, it doesn’t like to refactor when it should. These will continue improve as the model improves, and our code quality bar will go up even more as a result. My bet is that there will be no slopcopolypse because the model will become better at writing less sloppy code and at fixing existing code issues; I think 4.5 is already quite good at these and it will continue to get better. In the meantime, what helps is also having the model code review its code using a fresh context window; at Anthropic we use claude -p for this on every PR and it catches and fixes many issues. Overall your ideas very much resonate. Thanks again for sharing. ✌️
English