Harpinder Jot Singh

" Omg ! I started using Qordinate ! And it’s incredible.. I have been on search for something like this from the time ChatGPT came out with tasks feature.. I am making it a lifeOS and it’s incredible . Thank you bro ! " This is the kind of thing which keeps me moving @qordinate_ai Glad to make difference one step at a time.

@nikunj @xiz25 lol

@xiz25 Blocking reply bots is more fun than reverse engineering network requests

Reverse engineering APIs through network requests is one of the most fun things you can do with Claude Code to automate tasks.. SO many websites are impossible to navigate "deterministically" via the DOM (or through screenshots). So, I just point Claude Code to use browser_harness by @browser_use (or vanilla playwright) and ask it to sniff network requests on the pages that I'm trying to get info on. And, I just keep clicking around on the sections of data that I want. And, then Claude Code is generally able to go through the logs to figure out what is the right structure for these APIs and what kind of auth do they need (most are cookie based). We also determine what kind of rate limits exist based on trial and error. I'm able to use that to construct jobs that allow me to get that data programmatically. There are many use cases for this besides scraping. I use this for random side projects (like the travel CLI), for monitoring websites (for intel), and for many many other use cases. Every website will soon need to be headless, and we'll need to figure out mechanisms for how we have our agents pay these websites programmatically as well. Just as we have llms.txt for data and structure, we'll soon need tools.txt for agents to determine what tools exist that can be leveraged.

@nikunj i actually did something similar but i made chrome extension, which would scroll and capture things and output in structured format doing this for X, reddit, and linkedin and a UI on top right now using this to monitor web for my ICP and related discussions

@DanielSmidstrup Slap someone

Tell me one thing you can do that CLAUDE cannot do yet

@thepushkarp @karpathy This is the billion+ gpt images that are being created @sama

congratulations @karpathy!!!

@ankur_d Loved reading this ankur, wishing you the best for userlens.

@bensen @Microsoft @openclaw if always-on means agents constantly scanning inbox/calendar, that gets expensive and noisy fast. better shape imo is intent-based triggers: tell the system what the agent cares about, let a small event layer watch M365, and wake the agent only on matches.

"@Microsoft is piloting '#ClawPilot,' an always-on AI assistant built on the open-source @OpenClaw framework, with over 3,000 employees testing it internally. The system is designed to proactively manage tasks across #Microsoft365, from inbox triage to scheduling, without user prompts." msn.com/en-us/news/oth…

@alphabatcher i'd also split level 4 into cron vs wake rules: weekly reports can stay scheduled, but inbox scans / backup issues should probably wake only on a real match, with proof the agent actually ran. otherwise one VPS becomes five agents politely hallucinating uptime lol

One VPS. Five Hermes agents. One folder that says who can touch what. The setup: /root/vps-agents > docs > rules > runbooks > env map > agent registry > restart commands > backup notes > no raw secrets /srv//data > .env > config > SOUL.md > memories > skills > cron > sessions > logs Start in 4 levels: 1. One Hermes agent personal assistant, Telegram or CLI, real daily work 2. Direct specialists SEO, dev, CMO, outbound, life 3. Orchestrator + specialists one front door, routed tasks, summaries back to you 4. Scheduled team weekly reports, inbox scans, backup checks, content runs Add the orchestrator after the specialists prove they are worth keeping.

wild numbers. one thing i'd love to see in the architecture post: how much of Slash is human-pulled through Slack tags vs system-pushed from events like PR opened, alert fired, ticket created, deploy shipped? feels like that trigger/filter layer is where a lot of the reliability + inference cost tradeoff lives.

We recently built an AI assistant inside @Razorpay called Slash. It reads our entire codebase, debugs production incidents, reviews specs, writes code, reviews every single PR, answer tech queries and also raises PRs for small features. It's easily accessible through Slack. We can tag it in any Slack thread, describe the problem in English, and it gets to work. Six weeks ago, Slash handled 122 tasks in its first week. Last week it handled 14000+. Queries, analysis, bug fixes, PR reviews, test runs and work that earlier lived across scattered tools and teams can now be done with Slash right within Slack. 1000+ people used it in a single week because it got their work done faster. The whole adoption has been completely organic. The numbers from last week have been very encouraging - 14,854 tasks completed. 2,150 PRs raised, 1,152 merged, 45% of those PRs shipped with zero human rework. A payout gets stuck mid-retry during a live incident, an engineer tags Slash and within seconds, it cross-references logs with code and pinpoints a state machine bug blocking the retry-to-failed state transition. Tells the team exactly which logs to check and how to resolve the incident. With its K8s analyzer skill, Slash scanned a single namespace, right-sized all 11 workers using 48-hour P95 pod metrics, and raised the PR. One run saved $560/month. A marketing banner bug was fixed with few prompt iterations with a PR raised, merged to prod and deployed in minutes. No front-end developer touched the code. Security teams ran static security testing and remediation through Slash at org scale. Thousands of findings were purged and many more got validated autonomously. But Slash isn't just an engineering tool. Account managers now trace stuck customer payments and integration failures through Slash instead of pinging engineers on Slack. L2 product support tickets get triaged by Slash before they reach engineering. 250+ non-engineers ran thousands of sessions last week. PMs used it for research on our payments infra, customer interviews and product features sometimes raising PRs of their own. Analytics teams built SQL pipelines. 11% of all sessions came from people outside tech and product. On our company bakkar (watercooler) Slack thread, someone asked Slash jokingly to assign tasks to everyone and it responded in the same tone. It seamlessly started participating in inside jokes and conversations. The quality compounds with use. Engineers who shipped 11+ Slash PRs averaged a 63% merge rate without rework. First-timers averaged 37%. Across the org, human review comments per PR have dropped more than 40% with Slash starting to do in-depth review of every single change. We're still early. Large cross-repo refactors, fully agentic sdlc and plan mode are next. But Slash has already changed how people at Razorpay build, debug, and ship every day.

@andrewchen Building Watchline: an event layer for proactive agents. Developers or Agents define "watches" over any source, and we find signal from noise without expensive LLM polling. About me: AI infra eng, 4.5y workflow/agents platforms; OpenAI Hackathon 3rd, ICLR’26 KG retrieval.

@adrielyong @speedrun I built a personal assistant and hit the same wall while making it proactive: when should it wake up, and how do you avoid burning tokens on every event? Now I'm building the infra layer that lets agents subscribe to what they care about and receive only the events that matter.

been referring many folks to the @speedrun team over the last few weeks! Sorry if I missed some DMs, feel free to reply below to ping me on what you’re building someone also just made a directory with everyone associated with speedrun: noticed.so/speedrun great resource

@elonmusk deepwiki.com/xai-org/x-algo…

The latest 𝕏 algorithm has been published to GitHub github.com/xai-org/x-algo…

@snowmaker While applying to YC, I reached out to many alums for helping with a review of our application Many of them went out of their way to help even when they didn't have any reason to. Thanks @Andydy42 , @dexhorthy, and all others.

If you're wondering if you should move to SF, watch this video.

@aaliya_va @ycombinator thanks Aaliya

@singhhcoder @ycombinator congrats on the launch upvoted

my gdoc homescreen right now thanks to all @ycombinator alums and others who were so kind enough to review our application

I upgraded for you @sama

I've a genuine question: for codex desktop app, does it go through proper testing before releasing updates? After every update something seems unusable/broken from before. for eg: in latest build, the right scrollbar is not scrollable because of the branch details pop up taking over. before, the popup from + icon in command bar didn't come up unless you are at some level of zoom. I love codex, but these issues make using the app very hard tbh.

You can now keep codex going for days. With GPT-5.5 it will build an entire OS kernel for you if you ask, or find critical bugs in a codebase, or optimize your database schemas, or… the options are endless.

Best part about applying to @ycombinator ? It forces you to introspect, and articulate in simplest words what you are doing, why you are doing it, and for whom? This is the skill you anyways need to master as a founder to hire, and to sell your thing.

@garrytan So can you share your {SOUL,USER,AGENT}.md?

The secret to an articulate agent like mine isn't one file. It's three: SOUL.md — Who the agent IS. Voice, values, operating principles, what good output looks like, what bad output looks like. Not a system prompt, a constitution. Mine says things like "brevity is mandatory," "humor is mandatory," "never open with 'Great question,'" "swearing is allowed when it lands." The more specific and opinionated this is, the less your agent sounds like a chatbot. Write it like you're briefing your smartest friend on how to be you, not like you're configuring software. USER.md — Who YOU are. Not a bio — a deep model. How your mind works, what you're building, your strengths, your blind spots, your family, your temperament, what triggers you, what you care about. The more the agent understands about you, the better it can serve you. Mine is ~4000 words. AGENTS.md — Operational rules. What to check on every message, what to never do, how to handle failures, lookup chains, path rules, brain-first protocols. This is the playbook for how it works, not who it is. The articulation comes from SOUL.md being brutally specific about voice. Generic instructions → generic output. If you write "be helpful and concise" you get ChatGPT. If you write "speak like a peer with taste, one sentence when one sentence works, uncomfortable truths welcome if actually true, language with voltage" — you get something alive.

@lifeof_jer stop blaming others just because you fucked up. if you don't know what kind of tokens to keep and give access to, don't blame cursor or railway for your mistakes

@thejesonlee @travisk @bgurley It's a deep wound

Breaking: @travisk and @bgurley are still not friends