visiting partner @ycombinator || founder at @instantexperthq @doverhq @zincdotcom
San Francisco, CA
- Tweets 2,094
- Following 516
- Followers 4,983
- Likes 16,119
half the value of filling out a YC app is thinking deeply about your company and your unique insights
don't outsource that thinking to AI. it's easy to tell when there's lots of polish but little substance
I've read maybe ~10,000 YC applications. 8000 for YC admissions/PG quite a few years ago and anther 2000 informally as an alum.
There's a new and growing mistake and I want to share why it's more important than you think.
By the way - I try not to ever say YC admissions works this way or that way. That's YC's job and they're always changing things and making up their own minds.
But I will share my impression and you can see what you think - YC partners, please chime in as well if you like.
The YC application is fundamentally an evaluation of a founder's potential. And much less an evaluation of a specific business. (Airbedbreakfast transitioning into Airbnb because Brian, Nate, Joe never give up and never stop listening to their customers)
So, in your writing of the application, YOU need to shine through.
This batch, I read ~200 applications.
I'm guessing over half were significantly written, edited, or polished with AI.
That half was easily in the bottom half of my informal rankings, almost every time.
To be clear, most were NOT AI slop. I often couldn't tell at first they had used AI. Founders were pouring their heart and soul into their app - so I what I received was high quality.
But I think the AI really hurt them. Here's why:
Once polished by AI, I as reader lost insight into the personality of the writer.
I love the YC application because the questions are the same questions I would ask in office hours or casually to a friend that wanted help with their startup. It pushes the founder out of their pitchdeck/VC talk style into simple worded questions that can be answered simply as well. And great founders shine through when they just write like they talk. The LLM writing made answers feel more like a corporate cover letter, with the personality drained out.
The AI-assisted drafts also gave me a sense of mid work. The writing itself seemed like zero standard deviations from the mean. The LLMs had lopped off the outlier portions of sentences, paragraphs, and thoughts. Given the YC App is an outlier detector, a writing tool that lops off outlier writing is the opposite of what you want.
And the AI writing was needlessly verbose. Which hurts when a reliable predictor of great founders is that they're deep thinkers and deep thinking shines through in the density of their writing.
So...
I did an experiment. I asked a bunch of these seemingly AI-assisted app writers to talk with me live. I just asked them the same YC app questions and recorded their answers.
I then gave them back the transcripts. Unedited and unpolished.
And...
OMG! A good number turned from easy no to actual good applications.
Magic! Remove AI >> Improve application.
This happened BOTH with native English speakers and founders abroad.
So this is my recommendation to all of you.
Don't let AI touch your application, even for a second.
And maybe try the dictation approach with a friend.
My apologies to the YC admissions team if you already knew AI-written applications were crap and you were using this insight to help you easily move applications to the No pile.
YC's Winter Batch Application deadline is Nov 2nd.
6
3
98
5,207
it was mindblowing to see this actually work the first time i tried it
one of the coolest demos i've seen so far this batch.
We’ve raised $1.3M from YC & Afore Capital so you never have to type or speak again.
Typing is slow. Talking into your computer is awkward. Subvocal fixes both
Our glasses give you the speed of speech without you making a sound or touching a keyboard.
It tracks your internal speech muscles with non-contact sensing and decodes silent, almost invisible mouthing into text at more than 200 words/min.
Pre-orders are now live at:
subvocal.co/preorder?utm_sou…
2
1
31
2,766
one strategy for building a software business is to sell to startups while they're small and grow alongside the ones that succeed.
most startups don't become large companies -- meaning you're dependent on a small number of winners. you also need those winners to keep using your product (more and more) as they grow.
e.g. a small team might buy your infrastructure because they don't want to run it themselves. as they grow, you're charging more, and the customer might explore hiring somebody and building an internal version themselves.
if that happens, you lose the customers you were counting on to become your biggest accounts.
so i'd look for something almost every company needs, and that successful companies will keep buying more of. you need a broad enough pool of startups to find the winners, and a product they have more reasons to use as they grow.
6
3
18
1,018
if customers need to do something else before they can use your product, consider doing it for them.
e.g.
- if you're selling a way for startups to accept payments, it might be useful to help them incorporate so they can start taking money (stripe atlas!)
- if you're selling an AI support agent, you could turn a customer's past support replies into a basic help doc the agent can use.
- if you're selling software that tests which ads work best, you could help make the first few ads yourself, so the customer has something to test.
if customers want your product but can't get started, helping them through that first step might get you further than another feature.
8
1
43
2,113
my computer is constantly being driven by my ai agents.
if i had a whole team doing that, i'd be terrified.
if your employees use browser or computer-use agents, you should probably be using @graypassinc!
We're launching GrayPass (@ycombinator F26).
We help your software tell when an AI agent is acting on someone's behalf.
Before sensitive actions, we check who's in control, then allow, challenge, hold, or deny.
900K+ sessions processed. $50K in revenue this month.
@graypassinc
2
1
10
1,135
the length of a software pilot should be based on how long it takes a customer to see results from your product. a month is a pretty arbitrary default.
e.g. if you're testing whether software saves time on a task people do every day, you might get a useful answer in a week. if that's enough to decide whether to buy, you don't need to wait for day 30.
but if your software predicts whether a permit application will be approved, you may need to wait longer than a month for the decisions to come back. ending the pilot sooner won't tell you whether the predictions were right.
i'd agree with the customer on what result would be enough to make a decision, then work backward to the timeline. that can sometimes mean a shorter pilot or a longer one.
3
17
1,004
a tactic for finding leads that might want to use your software is to look at job postings: if a company is hiring for role that your product would replace, reach out to them and get them to use you instead.
but if there's a job posting live, it means you're likely too late to make the sale. by the time the role is posted, someone has made the case for headcount and secured the budget.
the posting tells you they did have the problem you're trying to solve for them, but it also tells you they've already decided to solve it by hiring.
instead, find opportunities where your software can *help* the new hire. they're already spending money on this problem. give them something that makes that investment pay off.
7
1
17
1,132
i'm honored
3
19
842
having a successful career can make the earliest stages of a startup harder.
you've done well at school, then at work, and now you have to put something in front of people who might say "i don't want this." or they'll try it and leave.
i've seen many people be afraid to put themselves out there because they feel like their standard is so high, which is the exact wrong attitude if you actually want to make progress.
making frequent "contact with reality" is sometimes disappointing, but oftentimes you'll be pleasantly surprised!
25
1
104
4,979
max retweeted
The number 1 regret of YC founders at the end of the batch is that they waited too long to launch. This is the impact it can have.
39
15
277
30,116
google will be a $10T company if it can fix the most basic use cases for gemini in the sidebar
5
24
1,623
startup idea: let agents verify a company's claims without exposing the private data behind them.
e.g. my agent is choosing a people-search API. it asks: “do you have 5 B2B customers in my revenue range who've made 1,000+ calls with a 90% hit rate?”
the vendor doesn't want to share its customer list or usage data. but it might be willing to prove the answer is yes.
the startup would connect to the vendor's systems and let agents verify specific answers without seeing the underlying records.
agents get evidence to choose a product. vendors get to prove it works without opening up their customer database.
20
56
4,632
it's okay not to worry about inference unit economics for your very first users.
i'd use the best models, even if serving those users costs more than they're paying you.
early on, the most important thing is whether your users are having an amazing experience. if a cheaper model can't do the work, they might leave before they see why they'd want the product.
you're spending a ton of time developing your product–it'd be a shame if the reason users didn't love it was because you were using less-than-frontier intelligence inside of it.
also, it's likely that the cost of that level of intelligence will come down quickly. and worst case you can always change models or pricing later.
you could still set a budget/usage cap to avoid extreme outliers. but for your first 10-100 users, getting them to love your product is more important than making money on each of them right away.
7
18
1,273
i think there are a lot of startup ideas in niches that are small today but could have 100x or 1000x more activity because of AI.
the thing to look for is something people would like to do, but usually don't bother with because it takes too much time or money.
e.g. imagine a company owes you a $10 refund. you might give up because finding the receipts and preparing the paperwork takes more time than it's worth.
if your agent can do much of that preparation, you might actually follow through. court fees and rules still matter, but the effort required to pursue the claim could fall quite a bit.
on the other side, the business would have more complaints to investigate, resolve or defend. a law firm specializing in that extra volume could be an interesting business.
that might look like a small niche today. but a lot of workflows might rip at the seams when they are done 100x more often, and there's a bunch of business opportunities around that.
12
31
2,360
who makes good launch videos with a fast turnaround?
a bunch of founders i'm working with either can't find a solid video agency, or they have one but they're slow/bad.
they'd love to spend $5 to $10k for a ~2 minute video. i think the ideal experience is an agency that spends an hour with the founders (interviewing and recording b-roll) and turns the finished product around in 1-2 days.
i'd love to send a bunch of business your way if you can do this well.
168
3
207
24,053
if you're a young founder and trying to get your first few customers, asking someone experienced in the industry to mentor you can be a good way to get a meeting.
older people often like being asked for their expertise. they may also be interested in hiring you or acquiring your company, and that's okay! you can lean into that interest without changing your goal of selling your product to them.
once you're talking, you can learn what they need and show how your product could help.
22
1
93
6,048
max retweeted
Today, Ramp raised $1.85B at a $60B valuation.
Seven years ago, Karim and I wrote our mission on a Post-it: “Save companies time and money.”
Our mission hasn't changed. It’s just gotten bigger.
172
142
2,433
697,620
two people at the same company can be very different customers.
this came up in office hours with a YC team building cloud infrastructure for robots. their bet is that future models will become too large to run on the robot itself.
but someone trying to get today's robot deployed may be perfectly happy with a smaller model running locally. buying infrastructure for a future model doesn't necessarily help them do their job.
a researcher at the same company might be working on exactly that future model.
i asked whether the team was getting in front of the people responsible for that longer time horizon. being at the right company isn't enough if you're talking to someone whose job is to solve a different problem this quarter.
it doesn't mean the research team will buy. but i'd want to have that conversation before concluding the whole company doesn't need it.
3
19
1,224