Lessons from 1,000+ YC startups: Resilience, tar pit ideas, pivoting, more | Dalton Caldwell (YC)

Dalton Caldwell, managing director and group partner at Y Combinator, has worked there for over 10 years across 21 batches. In this conversation Dalton shares the patterns he’s observed across more than 1,000 funded startups: simple pragmatic advice that founders tend to forget (“sell, make money”, “just don’t die”), tar pit ideas that draw founders in and trap them, the methodology behind successful pivots, how to think about TAM (mostly: don’t, not yet), why generic growth-hacking advice is actively harmful at the seed stage, and the request-for-startups for 2024 covering 20 categories YC wants to fund — including ERPs, open source, defence tech, manufacturing in America, better enterprise glue, and small fine-tuned models.

Caldwell / Lenny’s Podcast — Lessons from 1,000+ YC startups (tar pit ideas, pivoting, resilience)

Lenny Rachitsky interviews Dalton Caldwell (managing director and group partner at Y Combinator, 10+ years across 21 batches) on the Lenny’s Podcast interview series — published 18 April 2024, ~81 minutes. The wiki’s first deep evergreen-YC-partner anchor on the pre-current-AI-wave practitioner-wisdom layer of the YC corpus — predates the wiki’s spring 2026 YC partner-content burst (Tan / GStack / Hu / Stanford CS153 / etc.) by two years and provides the foundational vocabulary (tar pit ideas / just don’t die / find an incumbent with low NPS / customer-validation-first) the later episodes inherit and assume.

TL;DR

• Tar pit ideas as a named primitive (~23:48–27:00, the talk’s most-quoted concept): “By definition it is only a tar pit if it seems like it’s not. If it’s just a regular idea that is hard, that is not a tar pit. The weird aspect of what we call a tar pit idea is an idea that a lot of people come up with and that it seems like an unsolved problem and you get lots of positive feedback for. And you have a really good set of arguments that it’s a really good startup idea — and that’s different than a bad startup idea.” Canonical examples Caldwell names: friend-coordination apps (“people have been starting that startup since the ’90s”), music discovery (his own founder-era tar pit), Foursquare-clone location apps (Lenny’s own tar-pit experience with Local Mind).
• The pivot-as-methodical (not artful) doctrine (~14:00–22:00): “It’s actually there’s a lot more science than art.” The Zip / Procurement worked example as Caldwell’s canonical demonstration: he coached founder Rujul Zaparde at Zip toward the prompt “intentionally find large publicly-traded or PE-owned companies that are hated by customers — combined with software is horrible” → Zip discovered procurement, pulled customers off legacy procurement software, built a billion-dollar business across six pivots. The wiki’s clearest articulation of the large-incumbent-with-low-NPS-pivot template that Glasgow / Campfire and Letter AI and Luminai later instantiate operationally.
• The struggle-is-universal anchor (~38:00–42:00): “There’s a lot of founders that come this close to it all being over and through sheer will kind of just keep it going.” Caldwell estimates ~50% of YC founders hit “actually truly got down to very very hard situations” at some point. The Winter 17 batch and the brex story anecdote (~12:08–14:00) — paired with Danny Alberon’s question “what is wrong with our batch — everyone is struggling, nobody is doing well, what have we done wrong?” — is the wiki’s clearest base-rate articulation of founder-persistence-as-universal-experience.
• Don’t optimise for TAM at seed stage (~28:00–33:00, Caldwell’s load-bearing contrarian-corner claim, restated later in the lightning round): “Trying to be super pedantic about market size when it’s like a pre-seed company or someone applying to YC is not something I put a lot of thought on. What’s the TAM of the collectible Funko Pop industry? I don’t know, I don’t think it’s that big — but I wasn’t worried about it. That was like the last thing I was worried about.” Brex India case study: “the TAM of [credit cards in India in 2015] was tiny because no one was using credit cards in 2015 in India. So you had to believe that the size of the credit card industry in India would 100× — well, guess what happened.”
• Growth-hacking advice is actively harmful pre-product-market-fit (~1:05:45–1:08:00, Caldwell’s contrarian corner answer in the lightning round): “Growth and growth hacking and doing all this analytics AB testing stuff is a total waste of time for very early startups. A lot of the advice was catered towards later-stage companies. The seed-stage founders would consume all the later-stage advice and get really confused. The anti-pattern I see is there are lots of founders that are very familiar with your [Lenny’s] awesome work — which I recommend, I like it — but when you have no customers and you’re reading Lenny’s guides on how to set up split testing and how you did growth at Airbnb, oh man, that is so not helpful.”
• Customer-validation-first as final tactical advice (~1:11:11–1:12:25): “My tactical advice is start doing customer validation first versus building a PowerPoint deck versus trying to raise money. If you find people that are really excited and you do line up customers, that is a great green light that it is time to do a startup. With that, when does the building come in? Build it once you have some conviction that you’re like — oh, I think I would have a customer.”
• The 2024 request-for-startups as the wiki’s first inventory of YC’s stated 2024 hunger list (~52:00–55:30): 20 categories including ERPs (Caldwell-personally-authored — “I get so few applications on that and they’re usually pretty good” — Glasgow / Campfire is the worked-example response that emerged from this same prompt), open source companies, space companies (“several of the folks that are actually going to space right now that aren’t SpaceX are YC companies”), a way to end cancer, spatial computing, new defence technology, bringing manufacturing back to America, better enterprise glue (“the software to connect all these business systems is usually pretty brittle and janky”), small fine-tuned models as an alternative to gigantic generic ones.

What was actually ingested

The full ~81-minute transcript via the yt-dlp auto-captions fallback path (the youtube-transcript-skill Playwright path hit the documented panel-render timeout twice — at 30s and 60s). Custom VTT dedupe applied: kept only the new content line per cue (with hour-aware <\d\d:\d\d: timing-tag detection — the original filter missed all content past the 1-hour mark). 2,244 deduped ASR segments, full hour-by-hour coverage to 1:20:49. Stage cues, [__] expletive-redacted marker, and host/guest turn-changes retained verbatim. Full raw at raw/videos/2024-04-18-caldwell-lennys-podcast-lessons-1000-yc-startups-tarpit-ideas.md.

Substantive content

1. The opening list of YC-partner-mantras (~0:00–0:50)

The episode’s pre-roll establishes the catalogue Caldwell unpacks across the conversation: “simple pragmatic advice — sell, make money. One of my mantras is just don’t die. Being coached and being reminded of the fundamentals and basics puts you in the right mindset. You have this concept of tar pit ideas. Recently you put out a request for startups — 20 categories of ideas that YC wants to fund. People say you’re the king of the pivot. A good pivot is like going home — it’s warmer, it’s closer to something that you’re an expert at. There’s a lot of founders that come this close to it all being over, and through sheer will kind of just keep it going.” The five rhetorical anchors of the episode.

2. Tar pit ideas — the talk’s most-quoted concept (23:48–27:00)

Caldwell’s operational definition: “By definition it is only a tar pit if it seems like it’s not. If it’s just a regular idea that is hard, that is not a tar pit. The weird aspect of what we call a tar pit idea is an idea that a lot of people come up with and that it seems like an unsolved problem and you get lots of positive feedback for. And you have a really good set of arguments that it’s a really good startup idea — and that’s different than a bad startup idea. A bad startup idea is something obviously bad or something where you just can’t get any positive feedback on.”

Canonical examples:

• Friend-coordination apps: “Building an app to coordinate with your friends to decide where to go out at night or where to meet up with people. It’s coming from a good place — it’s a good idea if you ask your friend, ‘Hey, would you like an app for us to coordinate to hang out more so we can be friends?’ They’re like, ‘Yeah, I would love that.’ You’ll get all this positive feedback from the world, and man, people have been starting that startup since like the ’90s.”
• Music discovery (Caldwell’s own founder-era tar pit at his first startup): “Music startups are hard. Trying to be like ‘oh, we’re going to fix music discovery’ is those classic things where you can get lots of positive feedback and even get users to work on those things — but there are aspects of it that make it a very hard idea.”
• Foursquare-clone location apps (Lenny’s tar pit, with Local Mind): “I had this startup called Local Mind that allowed you to talk to people checked in in various locations around the city on Foursquare and Gowalla back in the day. Everyone when they used it, they’re like ‘holy [__], this is the most incredible thing I’ve ever seen — I could see what’s happening at this bar that I’m about to go to’ — and then they never use it again.”

The diagnostic insight: “Part of being a true tar pit is that you can get good initial validation.” Tar pits don’t fail the obvious tests; they fail the retention test. The wiki should carry this as the load-bearing diagnostic primitive for founder-idea-evaluation.

3. The pivot template — methodical, not artful (14:00–22:00 + the Zip worked example 18:00–22:00)

Caldwell’s contrarian view on the pivot journey: “It can be quite methodical. At the end of the day, when you’re especially building something for a business customer, when you’re building a B2B company, there’s actually a lot more science than art.” Two parallel threads at the beginning:

1. Mission/story alignment: “is there a mission or story that truly energizes me to work on a problem for a very long time?”
2. Customer-pain saturation across enough conversations to abstract the underlying generalisable problem.

The Zip / Procurement worked example (the wiki’s clearest articulation of the find-an-incumbent-with-low-NPS pivot template). Caldwell coached Rujul Zaparde (Zip co-founder, formerly Flight Car) through six pivots before they landed on procurement:

“My suggestion was to start by looking at what companies are publicly traded and/or owned by private equity that are large and that also are hated by customers — and to try to intentionally find where there’s a knowable big market with an incumbent, combined with the software is horrible. They basically found out about all this procurement software and what the state-of-the-art was — and that was basically the prompt.”

The wiki should treat this as the named template that the AI-native-ERP-vendor cluster ( Campfire / Vori / Luminai) instantiates. “Basically find a large incumbent with very low NPS and try to disrupt them.”

The pivot-mountain-vs-desert framing from a Michael Seibel conversation that Caldwell endorses (~21:23–22:10): “You move towards the mountains and the desert to find the gold of a new startup idea versus the middle of the city. You’re unlikely to find gold in the middle of San Francisco.” The contrarian-geography metaphor for founder-idea-selection.

4. The Winter 17 batch and the brex story (12:08–14:00)

Danny Alberon (a YC founder) asked Caldwell during a Winter 17 batch session: “What is wrong with our batch? Everyone is struggling, nobody is doing well — what have we done wrong?” Caldwell’s response: he told them about the brex story (Caldwell funded ~35–40 companies in his Winter 17 group). The wiki should treat this as the empirical anchor for the founder-resilience-as-base-rate thesis — paired with Caldwell’s later estimate that ~50% of YC founders hit very-hard situations at some point.

5. The TAM/market-sizing contrarian (28:00–33:00, restated in the lightning round)

Caldwell’s load-bearing position on TAM at seed stage: “Trying to be super pedantic about market size when it’s like a pre-seed company or someone applying to YC is not something I put a lot of thought on. The things I’m worried about — it’s like, hey, how do you get users, hey, how do you grow, like are you making something people want — those are the things I’m really worried about as opposed to ‘oh, I ran an Excel model and I’m worried this might not be a big enough TAM.’ That’s not at the top of my list.”

The Brex India case study: “The TAM of [credit cards in India in 2015] was tiny because no one was using credit cards in 2015 in India. So you had to believe that the size of the credit card industry in India would 100× — well, guess what happened. So I’m not saying that having a large market someday doesn’t matter — of course it does eventually — but trying to be super pedantic about market size when it’s a pre-seed company is not something I put a lot of thought on.”

The acknowledgement that other investors do care about TAM: “YC is unique in a lot of ways where you invest very early and you help people through this journey. A lot of investors are very focused on TAM, so you may find you’re getting turned down because they don’t think there’s a big enough market for you to build a big business.”

6. The struggle / persistence catalogue (38:00–42:00)

Caldwell’s base-rate estimate: “I mean, everyone goes through that, and again there’s gradiation. People that actually truly got down to very very hard situations — it’s still a high percentage, like maybe 50%.” The wiki’s clearest founder-vantage statement that founder-resilience-during-hardship is the modal experience, not the exceptional one.

The empathy-arc for the founder considering shutdown (~10:50–11:30): “If you’re having a really bad time, it’s no big deal — no one will remember that you shut down your company, probably in 10 years or 20 years. As long as you have integrity, as long as you’re an honest person, as long as you handle yourself well through good times and bad, people remember you fondly. We have such a short life — there’s only so many years we get to have our careers. Doing something that makes you miserable and the only reason you’re doing it is to avoid losing face — that seems like a pretty big opportunity cost on literally your life.”

7. The contrarian corner — growth-hacking is premature for seed (1:05:45–1:08:00)

Caldwell’s load-bearing claim about Lenny’s own content category: “Growth and growth hacking and doing all this analytics AB testing stuff is a total waste of time for very early startups. One of the weird things about having lots of startup advice on the internet — and this is one of the reasons we started making videos at YC — is a lot of the advice was catered towards later-stage companies. Oh, here’s how you set up your board, here’s how you motivate your sales team — it was all aimed at series A / series B founders, not for seed-stage founders.”

The anti-pattern: “There are lots of founders that are very familiar with your awesome work — which I really recommend, I like it — but when you have no customers and you’re reading Lenny’s guides on how to set up split testing and how you did growth at Airbnb, oh man, that is so dumb, that is so not helpful. You see this inclination away from getting a first customer, getting one customer and talking to that person, and instead they have like all this really complex growth-hacking theory.” The cross-channel meta-comment (Caldwell criticising premature application of Lenny’s own content on Lenny’s own channel) is itself a small information-diet worked example.

The big-tech failure mode: “This also happens if you worked in big tech where your product already has scale. If you work at Facebook and your job is to launch new little features — yeah, of course you should make heavy use of analytics and AB testing and split testing and feature flags. But when you have no users, what are you doing?“

8. The 2024 request-for-startups inventory (52:00–55:30)

Caldwell describes the request-for-startups as an information-diet intervention: “We’re trying to mix up some of the information diet about what kind of ideas people might be contemplating.” Selected categories named on stage:

• ERPs (Caldwell-authored): “I made one about ERPs because I get so few applications on that and they’re usually pretty good and I just would love to see more people look at that and learn about what ERPs are.” The wiki’s Campfire (May 2026, ~13 months later) is the worked-example response that emerged from this same prompt.
• Open source companies
• Space companies: “Several of the folks that are actually going to space right now that aren’t SpaceX are YC companies.”
• A way to end cancer (“no big deal”)
• Spatial computing
• New defence technology
• Bringing manufacturing back to America
• Better enterprise glue: “The software to connect all these business systems is usually pretty brittle and janky. There’s still a lot more room for improvement, and likely LLMs will improve, so we’ll probably be able to create better and better glue so all sorts of software systems can talk to each other.”
• Small fine-tuned models as an alternative to gigantic generic ones

The wiki should treat this list as the April 2024 baseline against which the AI-native-vendor cluster ingests (Letter AI / Momentic / Luminai / Glasgow / etc.) can be read — most of those companies map directly onto Caldwell’s stated 2024 hunger list.

9. The final tactical advice — customer-validation-first (1:11:11–1:12:25)

Caldwell’s closing tactical prescription, before the lightning round: “Get permission to talk to potential customers and try to pre-sell something before you write code and have those conversations. My tactical advice is start doing customer validation first versus building a PowerPoint deck versus trying to raise money — versus all these other things. If you find people that are really excited and you do line up customers, that is a great green light that it is time to do a startup.”

The when-do-you-build question: “Build it once you have some conviction that you’re like — oh, I think I would have a customer. I think at least one person would use this thing I want to build. At least one.”

10. Career signal — Caldwell’s pre-YC trajectory (briefly)

Caldwell mentions his early-2000s San Francisco peer-group included Zuckerberg, Reid Hoffman, Sam Altman, Elon Musk, Sean Parker (~75:39). Founded Mixed Media Labs / Picplz / App.net before YC. Joined YC ~2014, has been there 10+ years across 21 batches at the time of recording.

Linked entities and concepts

Entities directly named or substantively discussed in the source:

• Y Combinator — accelerator; Caldwell is managing director / group partner. Source-count bumps 10→11.
• Dalton Caldwell — promoted from Dangling to entity page (this is his second substantive source — first was Glasgow / Campfire May 2026 where he was named as YC partner-of-record but not the speaker; this is his first headlining-vantage source). Existing Dangling-list entry on Y Combinator page should be retired.
• Lenny Rachitsky — host. Existing Dangling first-mention from Schoening / Lenny’s Podcast (May 2026). This is the second substantive source citation — second-source-promotion candidate, but the surface area of Lenny Rachitsky as a wiki-relevant entity is modest (he’s the interviewer; what matters is the channel’s content), so the wiki may defer the promotion further if his vantage doesn’t add wiki-load-bearing content beyond the channel-entity (Lenny’s Podcast).
• Lenny’s Podcast — channel-entity (existing; source-count bumps 3→4).
• Rujul Zaparde — Dangling first mention (Zip co-founder; the operational worked example of the pivot template).
• Zip — Dangling first mention (procurement software; YC company; six-pivot success story).
• Brex — Dangling first mention (the Winter 17 batch anchor anecdote; also the Brex India TAM case study).
• Stripe — substantive cross-mention (~51:13, on Patrick Collison’s conviction-building arc).
• Michael Seibel — Dangling first mention (Caldwell endorses Seibel’s mountains-and-desert metaphor at 21:23).
• Scott Belsky — Dangling first mention (Caldwell pulls the more conviction over time vs less heuristic from Lenny’s episode with Belsky on when-to-pivot).
• Danny Alberon — Dangling first mention (YC founder who asked the what is wrong with our batch question).
• Pete Kazi — Dangling first mention (Caldwell endorses his book Founding Sales).

Concepts touched substantively:

• enterprise-ai-adoption — the find-an-incumbent-with-low-NPS pivot template + the better-enterprise-glue / ERPs / small-fine-tuned-models request-for-startups categories that map directly onto the wiki’s AI-native-vendor cluster. Source-count: +1.
• dynamic-capabilities — the pivot-as-methodical (not artful) doctrine + the customer-pain-saturation generalisation process is the founder-vantage operational instantiation of digital-sensing/identifying-needs at startup-formation scale. Source-count: +1.
• strategic-foresight — the don’t-optimise-for-TAM-at-seed-stage contrarian + the Brex India case study as worked example of why market-size pedantry mis-predicts which startups become large. Source-count: +1.

New concept page candidate (deferred to next ingest pass; flagged for promotion): tar-pit-ideas — Caldwell’s named primitive. Multiple wiki sources (Letter AI, Glasgow / Campfire, Hu / YC framework, Mittal / Yhangry) gesture at the anti-pattern without naming it; this source explicitly defines it and gives canonical examples. Recommend promotion to a standalone concept page on next ingest that touches the pivot / founder-idea-selection territory.

Dangling (single-source mentions, deferred): Rujul Zaparde, Zip, Brex, Michael Seibel, Scott Belsky, Danny Alberon, Pete Kazi, Mixed Media Labs, App.net.

Caveats

• 81-minute Lenny’s Podcast interview from April 2024 — predates the wiki’s spring-2026 AI-native-vendor corpus by two years and predates the December 2025 agentic engineering era phase change. Caldwell’s advice is evergreen YC-partner wisdom on the founder-formation-layer, not on AI-substrate-shift questions.
• The pivot / tar-pit framings Caldwell describes are not derived from a published research method; they are the practitioner-vantage observations of a single YC partner across his ~10-year tenure. Treat as practitioner-aggregated heuristic, not as empirical study.
• The 50% of founders hit very hard situations estimate is a self-reported partner observation, not a measured base rate.
• The Zip-pivot-was-coached-by-Caldwell-toward-the-procurement-prompt anecdote depends on Caldwell’s own recollection; Rujul Zaparde’s own framing of how Zip found procurement may differ.
• The request-for-startups 2024 list named on stage is a partial selection of the 20 categories — Caldwell names ~9 explicitly; the full list lives at a YC URL Lenny promises to link in show notes (not ingested).
• ASR transcription artefacts: timestamps may drift ±1–2 seconds from the original due to the VTT new-content-line dedupe method; expletive markers preserved as [__] per channel convention.
• yt-dlp n challenge solving failed warning during acquire — caption-only download still succeeded; video-format download (irrelevant for transcript) may have been degraded, but transcript fidelity is unaffected.