Building dubdub in public. Sold my last one (Aldun, healthcare). Flipkart and Udaan before that. I post about why people behave differently than what they say.
Bangalore
Joined June 2010
- Tweets3K
- Following330
- Followers670
- Likes3.1K
Pinned Tweet
“Most people are trying to come across as intelligent / competent”
Very powerful.
This is obvious when someone is *selling* themselves in situations like interviews or fundraising, but realising that it is true in most scenarios puts a different perspective.
everyone gossips but no one admits that they do.
recently got curious about why we like to gossip.
Interesting learnings stood out.
> Scalable way of staying in touch with your network which you can’t do 1-1. Highlights propagate faster through gossip.
> Reputation tracking for both right and wrong doings. Think of it like cheap policing construct.
> Status competition. Some use it as a tool to bring others down and feel good about themselves it.
what are other behaviours you’ve seen everyone do but not admit it ?
Credit to @curiouscartel for a post on gossip and rumour which got me intrigued.
What is the real definition of “deep-tech” here ?
Is predicting what stock to procure across thousands of dark stores, deep tech ?
Or, startup enabling large companies to purchase carbon credits, deep tech?
VCs are in the business of IRR and capital deployment at large will only follow its natural evolution.
My X feed is filled with househelp startups burning enormous amounts of capital to capture market share.
Meanwhile, I know multiple deep-tech startups generating significantly more revenue with far less burn, yet struggling to raise even a fraction of that capital.
To me, this is one of the biggest problems with our ecosystem.
We say we want to build products in India for the world, but we invest far less in the future than the US does.
We cannot become a deeptech powerhouse if our capital keeps chasing the present instead of funding the future.
YT premium has got to have the lowest subscription cancellation rates. It is delta 8 experience (h/t @kunalb11 ) and there is no way you can go back.
Has anyone here cancelled their YouTube Premium subscription ? I’d be surprised to find one.
genuinely conflicted about early career folks using claude/chatgpt for thinking work.
Hear me out.
the models today are probably better than the average "experienced" professional, and they've made intelligence available in seconds for almost any problem. BUT BUT, they don't really train you how to think, and early career is exactly the time when you need the reps.
when i think about the good leaders i've had, they never handed over the whole solution to a junior. they'd nudge, point in a direction, let you struggle a little first. an AI just answers, every time, immediately. and the judgement of what deserves depth versus what doesn't is something people build over years of doing the work. that's the part that worries me - the output looks senior, but the reps underneath aren't happening.
i've occasionally resorted to explicitly asking the team to spend some time with a problem on their own before turning to AI. what's the right balance, thoughts?
Very interesting opportunity. Check it out.
The only config i am dealing with today is the one I update to turn on the feature flag for new launch.
there are two kinds of consumer AI companies, and you can tell them apart by where the AI cost sits.
> type one : every message costs fresh inference. more usage, more cost. character ai is the clean case. users create the characters for free, millions chat with them, and the company pays for every reply.
> type two: AI does the expensive work once and the artifact gets served like media. PocketFM produces its audio series with AI - 93% of the catalog now - and every listener after that is free. content cost down 95%, a series that took two years, now ships in a month, and the company is EBITDA positive.
type one keeps its cost in the data plane. type two pushes it into the control plane.
type two wins on unit economics but it needs the right calls and a highly experiment-oriented mindset.
In @shreyas 's class, when you are evaluating, say quality of marketing email, you are refrained from judging something as 7/10 as it allows you to take non-pointed opinion. This data adds more evidence to that approach.
İnsanlara 1 ila 10 arasında bir sayı seçmelerini isterseniz genelde 7 seçerler. 1 seçmezler, 10 seçmezler. O sayılar yüksek sesle söylenmiştir.
Birden zihinlerinde saymaya başlarlar. Genelde ona yaklaşırken onları cevap vermeye zorlarsanız, mesela dokunursunuz, o zaman bir asal sayı olan 7'i seçerler. 10'a yakın bir sayı. Güzel bir sayı. İçleri rahat eder.
Bu sebepten insanlar iyi bir rastgele sayı üreteci değildir. Bir takım "meyilleri" vardır.
LLM'ler daha da kötü. Bu eğilimi (bias) alıp onu bir davranış kalıbı haline getiriyorlar.
Aldun got acquired a little while ago, but reading this made me really happy. we picked a hair-on-fire problem, hospital discharge stuck waiting on insurance approval, and built something people actually love. 4.8* rating is the biggest win for me :)
it's 2026, why are people still talking about leetcode and DSA for interviews.
rather ask people to explain one problem they solved and dig deep, you will understand far more.
Have been hiring across a lot of roles and few things keep separating the candidates who move forward from the ones who don't. sharing in case it helps someone preparing for interviews. all of it is relevant for any kind of role - ops, tech, product, marketing etc.
> How well can you explain what your current/past company does.
sounds basic, but use this opportunity to establish your understanding and clarity. if your company isn't a popular consumer brand, this is free real estate - two clear minutes on what it does, who pays for it and why. mid to senior folks, i expect this even at a well known company, and not the pitch deck version. your take. it tells me whether you understand the business you work for or not.
> How well can you explain a problem you've solved.
context first, then the problem, then what you did, then what changed because of it. again by establishing the right context, you can establish your understanding of the situation and why you did what you did. most people skip the context and jump straight to the solution, which tells me nothing. and the impact has to be both numbers and subjective as well.
> Can you hold a conversation on that same topic.
i will drill down. why was this problem worth solving, what choices did you make, what did you reject. if you actually did the work, 15 minutes of depth is easy. if you watched someone else do it, this is where it shows.
> Can you sell your spike without explicitly saying it.
nobody should have to tell me they have "attention to detail" or "ownership". craft your story so the spike is obvious by the end of it. if i have to take your word for it, it isn't landing.
is god’s karma system : one giant if-else file which nobody has refactored in 3000 years ?
if (day == TUESDAY && food == CHICKEN)
karmaPoints -= 100
saw someone scrolling reels during a movie. Seriously how bad can our attention spans get. that you came to watch a movie on large screen and get distracted during that.
Haha. Brilliant cat and mouse game.
They banned giving tip before the ride so they have inverted it with a higher pricing and allowing to bargain lower.
Supan Shah retweeted
Started a small experiment around football fandom less than 4 months ago.
Today, 100K+ members and 2.4K weekly contributions.
One thing we're learning: people don't just want to consume content. They want to debate, participate, compete and show off what they know.
Still early. Lots to learn.
flipkart big billion days sale is on right now and it's bringing back a ton of memories from the first one in 2014.
we had done flash sales before. xiaomi, motorola, a few other events. but this was a different BEAST.
we did 20x the revenue of a usual day. and we were live for ONLY 10 odd hours.
I was in the ads team and something funny happened. there were established brands paying to put banners on flipkart. banner click through rates were usually 1-2%, a tiny slice of our traffic. and they went down. they couldn't survive the load.
think about that. 1-2% of flipkart's traffic was enough to bring down large established company websites.
the other thing i remember is how close i was to it. as an engineer with 1 year of experience, i got to watch decisions like our CTO prioritising the customer order path over committed ad revenue, while everything was on fire. it was never kept behind a closed door. that's just how flipkart worked, and i didn't know how rare that was until later.