Tanmay Gujar

graphql

Google has an internal "let it break" essay about a hero engineer whose hard work ends up being a net negative (by masking the underlying issues). My manager sent me that essay when I was trying too hard to get the collective TensorFlow unit test suite green.

@basit_ayantunde + horrifyingly long test names

Easiest sign of AI-generated code: Excessive comments that are consistently title-cased.

One of the best ways to uplevel is to work with great talent. Not everyone has had that opportunity. AI is going to change that. Everyone has access. Looking forward to a year of massive upleveling for anyone who wants it.

cuDF graph #GTC2026

Jensen’s favorite slide!

I bought a iPhone case for 9$ and a screen protector for 6$, I have dropped it several times and it hasn’t broken

I call it the "Refrigerator Principle" Most organizational dysfunction exists because everyone assumes someone else has the authority to fix it, and the fastest path forward is often just pretending you have that authority and dealing with forgiveness rather than permission.

Our Megatron Core MoE technical report (88 pages!) is out - probably the most complete systems guide to training MoE at scale. Covers parallelism, memory/comm/compute optimizations, FP8/FP4, and best practices. 1,233 TFLOPS/GPU on GB300 for DeepSeek-V3. arxiv.org/abs/2603.07685

@cursor_ai fork chat is a great feature! I want to use a previous chat state a lot of times after going down bad rabbit holes

I find that codex models work much better with the OAI extension or CLI as opposed to using the same model in cursor

@iamaheron_ why? these are good hobbies

swear to god if another new person at my company lists “running”, “hiking”, and “traveling” as their hobbies i am going to go feral

@cursor_ai 1. Scrolling while plan doc is incomplete is buggy 2. Would be nice if I could comment on specific lines in plan mode like antigravity 3. Give feedback button in the editor does not work

Devcontainer workflow seems broken, chat doesn’t work inside devcontainer, git repo enumeration does not seem to work too

clicking update on @cursor_ai

why pilots say minor bumps on the way when they are not minor at all,, bro I was tensing up muscles I didn’t even know I had

@sundeep Yes, most of my investment is just NVDA

Did you beat the index this year?

just looks out the window for hours. I wish I could replicate his zen

I used them a lot before chat came out, and I am now forced to take them out of my writing to avoid people from thinking it is AI generated

Cursor feels like a company without product management. Just devs and designers running seemingly random experiments with little higher-level strategy. I use Cursor a lot, and it is becoming a very frustrating experience: — The interface is cluttered with tiny icons that get shuffled around every week — Are they building an agent or an editor? Pick one. This hybrid approach, where they pivot weekly, is frustrating — “Everything and the kitchen sink” isn't a great way to manage a product. A better approach is focusing on the core, and make that great instead of changing/adding useless things — Focus on stability. People use this for real work. Running into bugs and issues daily isn't the experience anyone wants Compare this to Claude Code: one text box, clear identity as an agent. Simple and focused. It's no wonder it's growing a lot.

oh you have a data warehouse? that's cool i live in a data mansion. it's on a data lake. we even have a data dog