I’ve built and founded many projects, but founding @3FdotVC was one of the most formative experiences for me. I incepted the idea in January and pitched it to several investors (MV, IITP IC, VisionOne, etc.) with a simple goal: build a VC entirely focused on students.

Rate the landing page we designed in 4 hours using @Google Gemini 3 Nothing fancy, just a minimal rebrand. #BuildInPublic

Steel did not make buildings taller by being smarter. It made scale possible by changing where the load lived. AI will need the same shift before the skyline really changes.

This is the direction we are exploring at Autum. Not adding more agents, but reorganizing how computer work is represented, monitored, and verified so organizations can actually absorb infinite minds without breaking.

What feels underexplored is that computer use is a control and coordination problem. Once work is decomposed into smaller, inspectable actions with shared state and feedback, supervision can move upward. Humans stop being glue and start becoming governors of the system.

This resonated because it correctly identifies that the limitation is not intelligence but structure. Most systems still ask a single model to reason, plan, act, & verify across fragmented software environments. That’s why context collapses & humans remain embedded in all loop.

भाग्यं परिश्रमिणाम्

YC request for founders

Insane toolkit for founders! I have done this mistake of working with low energy folks and regretting.

One of the finest piece of reading material about computer use. Very well written by @a16z team https://t.co/a4v7p70ke6

The best time to start up and fail is actually while being in college. - You got a cushion to fail and restart. - people are okay with you making mistakes. - Software will come at discounted rates or free If it fails, it’s a project.. it will succeeds, a billion dollar!

If you ask me one of my favourite llm infra project, it’s definitely @UnslothAI I literally helped kids from 1st year at IITP get hands Unsloth’s colab notebooks and they tuned their own model for capstone projects!

- be Manus - have two products in a16z top50 AI list - $0 - $100M ARR under 9 months - outperforming frontier model labs on key benchmarks this is the journey of how it happened (from my personal bias😎)

Jeff Bezos in a 2008 talk for the startup accelerator Y Combinator... One way to predict the future is to compare it to something similar from the past.

It’s not about being talented and static, it’s about being consistent and evolving. Congrats team 🙌

There's an uncomfortable truth about ultra-successful people. Steve Jobs, Elon Musk, Jeff Bezos — they all credit their success to luck. So Paul Graham developed a framework to make luck repeatable and predictable. Here's how to consistently get "lucky" like the world's elite:

You don’t lose because your product is bad. You lose because no one knows it exists.

@malpani @salkhanacademy Hi Anirudh sir, Built it 😀 Currently, 2 interfaces, 1 for students & 1 for teachers. Lots of room for customisations :) Waiting for your review.. Here's the demo: https://t.co/scQdYKNzAe

Thinking of AI models as a stackable pieces of tech is insane. Not necessarily all the parts of a work should be done by one single agent. n8n and agentskit is ig just the tip of the iceberg.

Can we build intelligence that uses virtual machines, servers, browsers, and computers exactly like humans do? We’re just getting started — excited for what’s ahead.

I’m starting an independent research initiative: AutoRel (Autonomy Research Lab) — focused on high-intensity research in autonomous machine use. AutoRel is a non-profit research lab exploring a simple question: