पाठशाला Pathshala · विचार Vichār, The idea · Lesson 06 · Start

Tarpit ideas and why they keep looking good

Some ideas have swallowed a hundred teams and will look fresh to the hundred-and-first. Learn the categories, the mechanism that makes them attractive, and the one condition under which they work.

Pathshala, The Founder Library · 11 October 2026 · 8 min read

There is a class of idea that arrives in every batch of every accelerator in the world, from a different team each time, described with the same enthusiasm and met with the same instant approval from everyone the founders ask. An app to find new things to do in your city. A better way to split expenses with friends. Groceries at your door in an hour. Each one has killed a hundred companies. Each one will look original to the hundred-and-first.

Y Combinator’s Dalton Caldwell and Michael Seibel gave this class a name in a 2022 episode of their podcast, tarpit ideas, and have returned to it since as the term spread and got misused, including a sequel on how it applies to current trends in AI. This lesson explains the mechanism, lists the categories with their Indian casualties, and states the one condition under which a tarpit idea becomes a company. The decision tree in the middle is that condition made walkable.

What a tarpit is, precisely

A tarpit is not a bad idea. A bad idea gets rejected and nobody builds it. A tarpit is an idea that seems good, that many teams have attempted, and that keeps failing for a structural reason the teams did not see because the early feedback was so warm. The image is the La Brea tar pits: the surface looks like a pool of water, animals walk in to drink, and the ones that sink attract predators who sink too. Matt Rickard’s summary of the YC discussion gives the examples the partners used: apps to discover new things, ways to bet on things, and rebuilding everything in Web3, with his own additions including the Craigslist killer and the universal chatbot of 2017.

Three features distinguish a tarpit from an ordinary difficult idea. The problem is real and widely felt, so it passes the first test every founder applies. The feedback is positive and effortless, because describing the idea costs the listener nothing and agreeing costs them nothing either. And the failures are structural, meaning the market would not pay, the economics never closed or retention never held, rather than executional, meaning a particular team ran out of money or fell out. An idea with those three features will attract a new team every eighteen months indefinitely.

Why they keep looking good

The problem is universal, so everyone validates it. Everyone eats out, travels, has friends and buys groceries. A founder who asks ten people whether finding a good restaurant is a problem gets ten yeses, and mistakes a universal mild irritation for a well. The [previous lesson](/library/problems-worth-a-decade) put a number on the difference: frequency times intensity. Tarpits score high on frequency and near zero on intensity, which is why the graph looks busy and the bank account does not.

The yes is free. Paul Graham’s name for ideas that sound plausible because they resemble things that exist is sitcom startup ideas: a social network for pet owners is the kind of thing a television writer would invent for a character to be working on, and nobody would object to it because nobody urgently needs it. The absence of objection is the trap. A real problem produces specific, slightly annoyed feedback about how the current solution fails. A tarpit produces agreement.

Survivorship hides the graveyard. Every tarpit category has one or two winners who are famous, and dozens of dead companies nobody has heard of, and the founder’s mind retrieves the winner. Hyperlocal grocery delivery has a handful of national names in 2026 and left dozens of companies in the ground between 2015 and 2017. The famous names are the argument for; the graveyard is the argument against; only one of them is visible from the founder’s desk.

Oversupply of founders. Because the idea is obvious and the yes is free, more teams enter than the market could ever support, so even the ones that would have worked in isolation drown in each other’s discounting. Caldwell’s rule of thumb in his conversation with Lenny Rachitsky is that the idea is receiving initial positive feedback that will lead to long-term trouble, and that the trouble arrives after the money is spent and the pivot has become expensive.

The categories, with the Indian graveyard

Discovery. Restaurants, events, things to do this weekend, people nearby. The problem is real and the solution competes with free: search, maps, Instagram and word of mouth, each improved by a company with a thousand times your budget. India produced a steady supply of city-discovery apps through the 2010s; few are still being updated.

A social network for a group. For students, for doctors, for a language, for a city. The group already has a network; it is called WhatsApp. The successes in Indian social, and there have been some, began with a content format the incumbents did not serve, not with a community that already existed.

Hyperlocal delivery of low-margin goods. The textbook Indian case. In 2015 a wave of grocery-delivery companies raised heavily; PepperTap alone had raised close to $50 million, including a $36 million round in September 2015, and planned 75 cities by March 2016. In February 2016 it shut six cities, its chief executive citing the investment climate and a decision to focus on depth rather than breadth, and by April it had closed. YourStory’s review of the year’s shutdowns put the cause plainly: rapid expansion, customer acquisition through discounts and, most of all, cash burned on logistics that left margins negative. The food-delivery version, TinyOwl, raised about $25 million and merged itself out of existence the same year for the same reasons.

Consumer travel planning, expense splitting, personal finance for everyone, marketplaces for services with no repeat purchase. Each has the same signature: a universal problem, a free yes, a product people use once and forget, and a monetisation that requires a density of transactions the market never supplies.

The list is not a ban. It is a list of categories where the burden of proof is reversed. In most categories a founder has to show why the idea will work. In these the founder has to show why it will work this time, which is a harder and more specific question.

The one condition

A tarpit idea works when the structural cause that killed the earlier teams has measurably changed, and the change favours the new entrant over an equally funded copy. Both halves matter. The first is the [why-now](/library/why-now-timing-argument) argument applied to a graveyard: not “the time is right” but “here is the constraint that killed them and here is the date and number showing it lifted”. The second is founder-market fit applied to the same graveyard: if the constraint has lifted for everyone, the category is about to be a capital race, and the question is whether you can win one.

Hyperlocal grocery is the Indian proof. The 2016 generation died of delivery density and discount-driven acquisition. The generation that returned from 2021 as quick commerce changed the model rather than the pitch: inventory in dark stores a few minutes from the customer instead of pickers in someone else’s shop, a basket built around margin rather than around staples, and a customer base that had spent five years learning to pay for everything by UPI. In the quarter to December 2025 Blinkit, with 2,027 dark stores, reported its first adjusted EBITDA profit of ₹4 crore on operating revenue of ₹12,256 crore, its chief executive attributing it to supply chain cost efficiencies, a shift toward long-tail categories and operating leverage. The margin is thin, the capital consumed getting there was enormous, and the category is now a race between a handful of the best-funded companies in the country. That is what the open window of a former tarpit looks like: possible, expensive and not for everyone.

Notice what did not change. The problem was identical in 2015 and 2021. The enthusiasm was identical. What changed was the constraint, and the teams that won were the ones that could name it.

A tarpit is not an idea that fails. It is an idea that fails for the same reason every time, to a team that did not look.

How to check in a weekend

Search the graveyard. Put the idea, in its own words and three synonyms, into two startup databases and one news archive. List every company that tried it, with founding year, funding and how it ended. If the list is long and the endings are similar, you are standing at the edge of the pool.

Read the post-mortems. Founders of failed companies write generously about why, and former employees talk. The point is to extract the cause in one sentence per company and then see whether the sentences match. Matching sentences is the definition of structural.

Ask three investors one question. Not whether they like the idea. How many times have they seen it, and what happened. Investors see the graveyard from above and will tell you, usually with relief, because they were about to have to say no politely.

Name the constraint and date its change. If you can write “they died of X; X changed on date D as shown by number N; the change favours us because of Y”, you have passed the tree. If any clause is missing, the gap is your next week’s work, not a reason to start building.

If you are already in one

The signs are familiar. Downloads without retention. Praise without payment. A growth chart that needs discounts to stay flat. Each new feature validated by the same enthusiastic people who did not pay for the last one. The honest move is the one the tree gives: name the structural cause, check whether anything about it has changed, and if nothing has, change what you can change, which is the customer, the business model or the wedge, while the problem stays. Keeping the problem and changing the approach is a pivot with a reason. Keeping the approach and changing the slogan is how a tarpit gets a hundred-and-second entrant.

A ritual before every idea

Before any idea gets more than a week of your time, run it through the tree above and write the path down: the answers, the graveyard list, the cause in one sentence, the constraint and its date. Keep the pages. Over a year you will have a dozen of them, and the pattern across your own rejected ideas is the most useful education in idea selection you will get. The first time you reach the narrow green verdict, you will also know exactly why, and that sentence is the first slide of the deck.

Nothing here is investment advice. The categories named are the ones with a public record; the point is not that they are impossible but that they have a history, and that a founder should read it before adding to it.