shobhit

People of pi. BIG NEWS. I've sold out. Let me know how you feel about this in the comments below. mariozechner.at/posts/2026-04-…

@ponnappa @srijan_mahajan Discipline is always aspired but never achieved

tangentially, i feel that most kinds of excellence demand their own unique discipline. over time I realised that in my mind I was equating 'discipline' with 'military discipline' and rejecting the claim 'excellence demands discipline'. but every form of art (for eg) has a discipline that is necessary to excel. some of these disciplines have v little in common with military discipline. and some disciplines come naturally to a subset of people. the discipline of an elite athlete vs the discipline of the commited rockstar. some people can naturally sustain it, some can't.

Discipline is overrated. People who look like they have superhuman amounts of it rarely grind through resistance, instead, they find things where not doing said thing feels worse than doing it. "Locking in" won't help you find that, rather, you'll have to sit with yourself long enough to figure out what you care about & most people would rather grind forever & try to fight resistance than face that question.

@debaghtk I am yet to get a hang of this Claude codex ping pong of implementation and reviews

@Sinister_Light implementation review phase on the same thread as above

my chota-sa gstack github.com/debaghtk/dotsk… still feels unreal that too ai agents can pair with each other and it only needs a markdown file

Anyone tried fetching Amazon order data for agents?

@SamBroner Congratulations Sam!

@soumilrathi Context

Starting a memory researcher group chat. If you’re into memory systems, personalization, or context engineering in LLMs & AI systems. comment “context” to join.

@thdxr Should be

ai password manager is that anything?

@threepointone Need some security first primitive on agent networking

what if this is only part of the story Think about it

@vaibhaw_vipul On the other hand, as a programmer we always craved to be at the frontier. And then somehow we all reached the frontier, and books will come a few years later. This is how frontier feels like?

I am not being a purist but honestly I don't understand this world now. I grew up in a world where skill deprecation was slow and predictable. Get mastery in X, mastery gives security and it compounds overtime. I am not sure how to operate or what to optimize for over long-time horizon. There are no mental models available to me as a programmer to navigate this uncertainty. Long-term optimization assumed a relatively stable skill market. That assumption is broken. The discomfort or hollowness I am feeling is not about economic value alone. It is about identity. Being a programmer for me was special - a way of thinking, a relationship with machines, a kind of fluency that was hard earned. (I am not gatekeeping, just stating my admiration towards legend programmers). The worse problem is that the alternative mental models are cope - "creativity", "abstraction", "move up the stack". Why should I pick up a new book? Why should I dive deeper? I don't know , this may feel like a rant because it is one. I have no conclusions.

@threepointone Auto harness

coming up: the meta harness

@blennon_ @rajivist @clairevo @Apple @Google Can I 🙋

@rajivist @clairevo @Apple @Google want to try this today? DM me

I’ve been thinking a lot about what would be a killer version of openclaw: > collaborative with entire family > shared context/memory > smart home voice assistant > available in phone too > can do all the below that @clairevo mentioned @Apple and @Google best positioned imo

Look at that slick TUI

I am so proud of what we built, and what we say on public record. Hopefully people will understand that what it takes to build such a strong culture. I hope that this clarifies many assumptions and allegations about @nadiemmakarim Google's posted a blog on similar lines. #gojek4life @gojekindonesia gotocompany.com/en/news/press/…

Changing economics of software building . What qualified as asset can now be liability.

@championswimmer @openclaw Object capabilities based auth model makes sense

The one thing @openclaw painfully makes clear is that we need an evolution of "authorization" models for agents. Currently there are two schools of thoughts on how to run Claw 1. Make it own accounts for everything including GMail, Whatsapp and everything 2. Run in on a PC where you are already logged in to everything so that it can inherit your tokens Almost all knowledgeable people of this space like @simonw have clearly advocated for sandboxing and have admitted they would never run it with full access to their life. Even @steipete recommends giving it aleast a separate number. But if you also make it a separate Google account, you now hobble a lot of its ability to handle your calendar, see your favorite places on Maps or summarise your emails for you. What doesn't exist (well technically does via auth "scopes") is giving Openclaw read only access to your Gmail without the ability to write. I think we need to evolve classic Oauth2.0 into another iteration where we acknowledge the existence of agents working on behalf of us. Just a reminder, Oauth2.0 already acknowledges the existence of "clients". In fact when you use Oauth2.0 you basically do the equivalent of giving your mobile app or the webapp your are using access to your account. Within that framework, we might probably need finer grained difference between when an Oauth token is being used for a fully automated LLM-at-wheels flow and when it is being used to give access to an app or website that a human is using. And then some sane defaults of what for-human tokens have access to and what for-agent tokens do.

@vedang @_svs_ Next week afternoon works. Just announce based on your availability, will show up

@Sinister_Light @_svs_ This weekend is really packed for me folks. Happy to do it sometime in the next week, per your conveniences. (A slot in the afternoon, between 2 to 4 pm, works best for me)

Claude Code multiple instances were eating a lot of memory so it built itself a queue on the filesystem and now it makes sure it only parallelises as much as the server will accept.....

@_svs_ @vedang Yes - Pi is on my review list too, @vedang please host

@vedang got a business to run...going mad chasing the new new thing. i am at good enough....but i am very very intrigued...let's hang on the discord this weekend you can show me around?

When people say “SaaS is dead”, they are too focused on software and very little on service - while in reality the other S matters more.

@lcbisu It’s not true - every code can be worked with AI. I have personally stop being a writer, now mostly an editor

I am hearing this again and again, from our team and from other companies too.. If you are starting from scratch today, you can go almost fully AI first. AI can write most of the code now. But if you are an older company with a large complex codebase from the pre–genAI era, it is harder to use AI generated code. You can still use it, but the contribution is much smaller compared to hype in the market. are there any solutions to make AI code writing work well for older, complex codebases? Has anyone figured this out?