nipungkant's profile picture. GM at @doordash. Formerly at @scale_ai, founder of Haven (@ycombinator W20).

Nipun Kant

@nipungkant

GM at @doordash. Formerly at @scale_ai, founder of Haven (@ycombinator W20).

One of the most important things to appreciate in larger companies is that adding more resources / people to projects often makes them slower Not faster


the courage to define what you won't do, is underrated

Incorrect: Good strategy is about how we will be 10X better. More correct: Good strategy is about where we will choose to be 10X better (differentiate), where we will be 1X (meet table stakes), where we will be 0.1X (below table stakes), and the rationale for these trade-offs.



2 failure modes of product management in complex orgs: 1/ Focus on metrics before vision & strategy 2/ Reliance on one metric as litmus test for PMF & demand Instead: start with vision then articulate strategy then lay out stretch goals then work backwards to get there


Very true

What no one is talking about is the fact that as soon as a new model drops, the agent harnesses aren't optimized for it. There's a ridiculous amount of elbow grease that you have to put in to actually make new models work truly well with agents. We're working really hard at…



“It’s a game of inches” is one of the most pernicious myths in tech companies It invites incrementalism that stifles the ingenuity needed for 10x gains


Highly recommend @crmiller1’s Chip War Striking how much it rhymes w/ the LLM era — the gold rush, obsessive competition, 3P interdependencies, and most of all, the once-uncertain market sizes that a few bet on


Once you start building voice agents, you’ll start to see possibilities all over the place There is no going back


Plenty of clickbait about vibe coding But to actually get apps to production, you must be able to understand.. -Your codebase -High-level architecture -How the agent intends to implement And be able to course correct it often


If you’re a consultant or mba looking to break into startups, the best thing you can do today is build something. Build a piece of software that solves a real problem for 1 person That’s more possible for anyone today than ever before


Much of building AI agents today is grasping for straws of determinism In a newly non-deterministic world


Tech jobs have trended toward specialization over past decade That is falling away rapidly as people have PhD experts in their pocket High agency generalists who can also code may be the highest leverage role


If you're vibe coding large features, best results come from 80% planning, 20% execution. Allocating way more tokens to the planning phase is the biggest bang for your buck.


Long live the app layer! H/t @Suhail

It might turn out that the GPT wrapper has more of a moat than GPT.



Painful annual planning comes from a lack of aligned strategy H/t @shreyas


o1-preview is a great step forward for independent reasoning, but is RLHF’ed to be a bit too verbose and detailed I hope @OpenAI tweaks in their next iteration


Nipun Kant reposteó

Are you thinking big enough?


Founding a company, over years, shows you the core essence of your being.

This is just so, so good. I highly recommend all founders read it... "Dying (no matter the size of the death) is something we humans tend to avoid, so for most people, this process takes decades. Well into middle or old age. For many, they never figure this out, and die clinging…



All are born original, sadly most die copies Create the life and career you want for yourself. Don’t passively let it be defined for you.


Yes, lots of growth from OpenAI et al But the avg worker at a big tech company is very far from truly incorporating AI into their daily workflows


+1, huge fan of @jack_moshkovich

After it came out that I’m supporting Trump this election, it was no surprise to be attacked by lefty journalists on multiple topics. The far-left Rolling Stone’s hit piece on me a few days ago based on our policy org Cicero’s success - we’re helping several states reduce…



United States Tendencias

Loading...

Something went wrong.


Something went wrong.