You got rejected from Y Combinator. Now what?

Disclaimer first, because it matters for how much weight you give this: I wasn't in the room. Everything below is other people's receipts — a founder who got into F26 and wrote down all 25 questions they were asked, someone who used to read applications, and a couple of people who have scaled things I haven't. I collect these. Here's what's in the notebook.

The ten minutes are almost entirely about traction #

The 25 questions break into three buckets and basically nothing else: traction, customers, founders.

Partner 1:

• Tell us what you're working on.
• What's the current status? What's built, how does it work?
• What does it do, how does it work? We saw the demo, talk us through it.
• What problem does it solve, for who?
• So this is software that augments the control room for large energy companies?
• So the labour problem is knowledge workers they can't hire?
• Talk to us about your customers. Who are they, how do they use it, how much do they pay you?
• Is the per-agent price an annual recurring contract?
• There's a fixed onboarding or setup fee too, right?
• What's your annual recurring revenue right now?
• Is that all from one customer?
• Who are the new customers and how did you get them?
• What's the actual activity of your agents on their systems? What's happening for them daily?
• Does anyone at the company need to interface with the agents, or is it largely autonomous?
• Backstory. How did you two come together, why are you doing this together?
• How did you pick this idea? You don't have a background in this at all.
• You've raised money already. When, and from who?
• You have a team now? Tell us about your team.
• You manage the engineering team?
• What's the plan going forward? Europe or US? Where will the company be based?
• Where will your sales energy be focused over the next three months?
• US or Europe?

Partner 2:

• This feels like a forward-deployed motion. How much software are you actually building that's part of the platform?
• What's special about the data standardisation you do?
• What's the defensibility? You're replacing a human, so you charge a lot. What stops someone coming in and repricing you by product?

Roughly half of that is commercial mechanics. Who pays, how much, is it recurring, is there a setup fee, what's ARR, is it all one customer, how did you get the new ones. The product questions are surface-level and get resolved in four or five exchanges — they saw the demo, they're being polite. The only hard strategic questions land at the very end, from the second partner.

Which is uncomfortable if you spent three weeks polishing the product story: it's table stakes and it buys you nothing. The time goes on revenue composition, acquisition path, and whether you have a defensibility answer that isn't "we'll move faster than them".

The filter is slope, not size #

There's no revenue or usage minimum. What gets read is velocity, measured in roughly two-week units. Per cycle: did you ship something significant, or did you learn enough to change direction? And with customers, not alone in a room.

The case that makes it concrete: a founder with about 100 customers was rejected after an interview. Reapplied three months later — same vision, better answers, 2x revenue, 2x customers. In.

Traction wasn't the problem the first time. A hundred paying customers is traction. What was missing was a demonstrated slope.

After a rejection, don't optimise for the next application #

The reversal an ex-application-reader uses, and it's the most useful thing in these notes: assume you're never getting in. Actually accept it. Now — what should the startup do in the next two weeks?

The answer is almost never "build more". Building is the comfortable excuse. It feels like work, it's entirely under your control, and nobody can reject you while you're doing it. What comes out of the question instead is: launch earlier than feels ready, sign a design partner, get the first euro in, go and sit with customers in person, and be able to say what you learned that you didn't know before and what data shows anyone wants this.

Same for the application itself — read the questions again, find the two where you were vague, and fix the thing underneath rather than the wording.

The convenient part is that "what do I do now" and "how do I get in next time" have the same answer. Twelve weeks to the next deadline is six two-week cycles. Six chances to have a different story, none of which require anyone's permission.

Ten customers who'd be devastated #

Not 1,000 signups. Not 100 people who like you. Ten who'd be genuinely upset if you switched it off tomorrow.

Ten is enough to tell you what the real problem is, what the product should be, and where to find the next ten. A thousand signups tell you your landing page works. Same instinct as everything above — depth of customer contact over breadth of metrics.

Leverage is in subtraction #

Almost all advice is additive: more hires, more process, more features, more meetings. The returns are usually in the other direction — removing a dependency, automating the manual thing, killing a process nobody remembers agreeing to.

Complexity is trivially easy to create. Simplification is real work and there's no dopamine at the end of it. I'm biased here, I build tools that delete things for a living.

From the same notebook #

Two things with nothing to do with YC that came out of the same conversations, so, while we're here:

Don't self-host email. From someone who scaled pipelines past 45B a month: the domain has too much undocumented, ambiguous, out-of-date surface area for an LLM to get right, so "I'll have Claude build it" fails in ways you won't see coming until your deliverability is already dead.

Positioning arbitrage is currently absurd. A YC startup appended "…for data centers" to its directory description and reported 25 investor inbounds the next day. Around the same time OpenAI launched a $400M early-stage AI fund. Both point the same way: capital is chasing AI-adjacent framing hard enough that the framing itself is doing measurable work. Useful to know, extremely easy to over-rotate on.

The one line #

They're not measuring where you are. They're measuring how fast you got there and whether customers were in the room while it happened.

Anyway. Twelve weeks, six cycles. Go find the ten people who'd be devastated.

Reapplying and want a second pair of eyes on the answers? Hit me.