Pick One Idea and Go Deep — Transcript

Full transcript

1. 0:01[music]
2. 0:09>> Hi, I'm John and I'm a partner at YC. I
3. 0:12often meet founders who have lots of
4. 0:14ideas about what to work on and can't
5. 0:16decide between them.
6. 0:17Sometimes they're working on multiple
7. 0:19things.
8. 0:20Often they'll say that they're waiting
9. 0:22to find the best idea before fully
10. 0:24committing, but it's extremely hard to
11. 0:26make meaningful progress on a startup
12. 0:29without committing to a single idea.
13. 0:32So in this video, I'm going to give you
14. 0:35a rubric for how to stop overthinking,
15. 0:37pick an idea, commit to it, and then
16. 0:41figure out fast whether it's actually
17. 0:43working. The most important piece of
18. 0:45advice I'd give to founders struggling
19. 0:47to pick a startup idea is
20. 0:49don't overthink it. Overthinking a
21. 0:51startup in the earliest days can take
22. 0:54many forms, but here a couple of the
23. 0:56most common failure modes I see. The
24. 0:58first is thinking that you need to find
25. 1:00the perfect idea. In some ways, this is
26. 1:03an understandable impulse.
27. 1:05Startups are hard, so shouldn't you
28. 1:07figure out the best idea before
29. 1:09committing? The problem with this
30. 1:10approach is that it's impossible to
31. 1:12figure out the perfect idea in the
32. 1:15abstract. You can only figure out what
33. 1:17you should be working on by making
34. 1:19contact with reality and getting
35. 1:22feedback from customers. The second
36. 1:24overthink is am I the perfect founder
37. 1:27for this?
38. 1:28It's true that founder market fit
39. 1:29matters. A non-technical founder likely
40. 1:32won't be the right person to come up
41. 1:34with a killer dev tool startup idea, for
42. 1:36example. But often founders, especially
43. 1:39second-time founders, weaponize this
44. 1:41line against themselves. They convince
45. 1:44themselves that they need a decade of
46. 1:46domain experience before they can start.
47. 1:48The truth is, you don't. If you pick an
48. 1:51idea you're curious about, go extremely
49. 1:53deep,
50. 1:55and most importantly talk to customers.
51. 1:57It's often possible to develop
52. 1:59extraordinary knowledge in a short
53. 2:01amount of time. We see incredible
54. 2:03examples of this all the time at YC.
55. 2:06Take Blake Scholl, the CEO of Boom
56. 2:09Supersonic. Blake spent his early career
57. 2:12working on ad tech at companies like
58. 2:14Amazon and Groupon before deciding to
59. 2:17work on commercializing supersonic
60. 2:20flight. Lots of people probably thought
61. 2:22he was crazy.
62. 2:23But now Boom is a billion-dollar
63. 2:25company. So don't let the question of
64. 2:27whether you're allowed to work on
65. 2:29something stop you from starting. Once
66. 2:31you've stopped overthinking your ideas,
67. 2:34it's important to commit to just one.
68. 2:37Often, I meet founders who are working
69. 2:39on multiple ideas at once because they
70. 2:41believe that this is the best way to
71. 2:43figure out which one will actually work.
72. 2:45There are a couple problems with this
73. 2:46approach. The most serious is that it
74. 2:49tends to produce bad data. If you don't
75. 2:52actually go deep on an idea, but instead
76. 2:55juggle it with several others, you won't
77. 2:57get good signal about whether what
78. 3:00you're doing actually works. And if you
79. 3:02don't get good signal, then you could
80. 3:04either prematurely talk yourself out of
81. 3:06a good idea or convince yourself that a
82. 3:09bad one is worth continuing. The
83. 3:11solution to this
84. 3:13is to go in depth first. If you're
85. 3:15trying to decide between several ideas,
86. 3:17all of which look equally attractive,
87. 3:20pick one idea and go deep on it. What do
88. 3:22I mean by going deep?
89. 3:24The first thing is that you should burn
90. 3:26the other boats. That is, you should
91. 3:28explicitly foreclose your other startup
92. 3:31idea options. Stop working on them. Tell
93. 3:34any customers that you've pivoted and
94. 3:36work with single-minded focus on the
95. 3:39idea you've chosen. One way to think
96. 3:41about going deep is that it should feel
97. 3:43like wearing a new skin.
98. 3:45You should become an almost
99. 3:46unrecognizable version of yourself. This
100. 3:50could mean changing your company's name,
101. 3:52your emails, your website, and even your
102. 3:55internal narrative about why you're
103. 3:57building a startup in the first place.
104. 3:58For example, I worked with a startup
105. 4:01called GovDash that helps customers win
106. 4:03government contracts. They pivoted at
107. 4:06least five times before finding this
108. 4:08idea. And each time they explored
109. 4:10something new, they changed their
110. 4:12company name and how they talked about
111. 4:14their mission. At one point, I forgot
112. 4:17how to get in touch with them because
113. 4:18they changed their email addresses with
114. 4:20each pivot. By truly becoming domain
115. 4:23experts in government procurement, their
116. 4:25fifth idea worked so well that they
117. 4:27could barely keep up with demand. They
118. 4:29recently raised a Series B to scale the
119. 4:32business and meet that demand. Once
120. 4:34you've decided to fully commit to an
121. 4:36idea and go deep, how do you know if
122. 4:38you're actually doing it well?
123. 4:40The high watermark I use to help
124. 4:42founders answer this question is could
125. 4:44you actually run your customer's
126. 4:46business? Say you want to build voice
127. 4:48customer service agents for cleaning
128. 4:50services. The question isn't just
129. 4:52whether you've talked to 20 owners. The
130. 4:54question is if I dropped you into a
131. 4:56cleaning business tomorrow, would you
132. 4:59know how to run it? Do you know what
133. 5:02their daily crises are? Do you know
134. 5:04whether answering the phone is a top
135. 5:06five problem?
136. 5:08Do you know how much business they lose
137. 5:10when a call goes unanswered and what
138. 5:12they would actually pay to never lose
139. 5:14another one? These are the kinds of
140. 5:16questions you need to be able to answer
141. 5:18with very high confidence.
142. 5:20Another way to think about this is could
143. 5:22you teach a class on the problem you're
144. 5:24solving? Are you one of the most
145. 5:26informed people in the world on the
146. 5:28subject? Getting to this level will
147. 5:30involve lots of conversations with
148. 5:32customers and sometimes even literally
149. 5:35doing the job yourself.
150. 5:37But don't obsess over needing to talk to
151. 5:39hundreds of customers before writing
152. 5:41code.
153. 5:42The goal is to do both at the same time
154. 5:45in a tight loop. Deep understanding of
155. 5:48customer needs, then product delivery,
156. 5:50then deeper understanding of customer
157. 5:52needs,
158. 5:53then better product delivery. Real
159. 5:55customers using your product produces
160. 5:57concrete data that will complement your
161. 6:00abstract knowledge, giving you a sense
162. 6:02of whether what you're building is
163. 6:04actually working. Once you're going deep
164. 6:07on an idea, there's several ways to
165. 6:09validate whether it's worth continuing
166. 6:11to work on. The most obvious one is pull
167. 6:14from customers, but there's several
168. 6:16other qualities of good ideas in the AI
169. 6:18era that you should look out for as you
170. 6:21go. The first is that the idea sits at
171. 6:24the edge of what models can do today.
172. 6:27This might mean that your product barely
173. 6:29works on today's frontier models, but
174. 6:31will clearly improve as they get better.
175. 6:34You should understand the bottlenecks
176. 6:36impeding your product's performance
177. 6:38intimately. If a particular bottleneck
178. 6:41doesn't clear the way you hoped,
179. 6:43solving that might become the company.
180. 6:46This is a version of Paul Graham's
181. 6:47well-known quote that you should live in
182. 6:49the future and then build what's
183. 6:51missing. The second quality of a good
184. 6:53idea is that it should verticalize. By
185. 6:56this I mean that it should ultimately
186. 6:58sell an outcome.
187. 6:59For example, providing insurance or
188. 7:01medical care rather than just software.
189. 7:04In the AI era, the cost of producing
190. 7:07software is going to zero. So, the
191. 7:09things that actually become valuable
192. 7:11aren't just software for X. They're
193. 7:14customer trust, licenses, regulatory
194. 7:18permission, and outcome ownership. So,
195. 7:20if you want to get into the insurance
196. 7:22space, don't build software for
197. 7:24insurance companies, just be the
198. 7:26insurer. Similarly, rather than selling
199. 7:29back office software for banks, just be
200. 7:31the bank. One example of this is Corgi
201. 7:34Insurance, an AI-powered commercial
202. 7:36insurance company from YC's Summer 24
203. 7:39batch. They They not content with being
204. 7:41a tech-enabled broker or even a managing
205. 7:44general agent because that was just
206. 7:46owning a part of the solution. Instead,
207. 7:49they set an ambitious goal of owning
208. 7:51everything from underwriting to
209. 7:53providing customer service, the entire
210. 7:55commercial insurance stack, and even
211. 7:57took the unprecedented step of acquiring
212. 8:00an insurance carrier during their YC
213. 8:03batch to make it happen. Being the
214. 8:04full-stack insurance company allows
215. 8:07Corgi to underwrite any insurance line
216. 8:09in any vertical with a fraction of the
217. 8:12head count of traditional carriers.
218. 8:15They can offer far better pricing, much
219. 8:17faster turnaround, and own all of the
220. 8:20economics. That brings me to the third
221. 8:22quality of a good idea.
222. 8:24It should be the most ambitious version
223. 8:26of itself. It may seem unintuitive, but
224. 8:29the cost of pursuing a wildly ambitious
225. 8:32startup idea and the cost of pursuing a
226. 8:34modest one are roughly the same.
227. 8:37They're both extremely hard. They both
228. 8:39place extreme demands on your time. So,
229. 8:42aim at the version that, if it works,
230. 8:45rewrites a sector of the economy.
231. 8:48Because that's also the version that
232. 8:49protects you from competitors, attracts
233. 8:52the best talent, and has a moat worth
234. 8:54building. This could mean building and
235. 8:56selling into the most regulated
236. 8:58industries, like legal, healthcare, or
237. 9:01financial services, or taking on very
238. 9:04large incumbents, like a $10 billion
239. 9:07legacy SaaS company, or building hard
240. 9:09tech, like robotics for space assembly.
241. 9:12Now, what if you do all of this and the
242. 9:15idea fails? The good news is that you'll
243. 9:18be in a dramatically better position
244. 9:20than where you started. First, you have
245. 9:23unambiguous customer data. You know
246. 9:26whether there's actually a hair on fire
247. 9:27problem in this space or whether you
248. 9:30just talked yourself into thinking there
249. 9:31was. You'll have real conviction to base
250. 9:35a pivot on
251. 9:36and a better sense of how to execute
252. 9:38going forward. But more importantly, you
253. 9:41will often come away from the process
254. 9:43with a new idea that will actually work.
255. 9:46When most founders begin, they're
256. 9:47solving surface-level pain points. The
257. 9:50real opportunities are almost always the
258. 9:53deeper structural problems. In other
259. 9:55words, going deep isn't primarily a
260. 9:57process for validating the idea you
261. 9:59started with.
262. 10:01It's a way to find the better idea
263. 10:02underneath. This almost always happens,
264. 10:05especially if you're at the forefront of
265. 10:07what models can do today. You'll notice
266. 10:10the bottlenecks, the gaps, the dev tools
267. 10:13nobody's built. And one of those could
268. 10:16turn out to be the actual company.
269. 10:18Here's what I want you to take away from
270. 10:20this video. First, stop trying to find
271. 10:22the perfect idea.
272. 10:24Just pick one. Then, burn the other
273. 10:26boats.
274. 10:27Learn everything you can about the
275. 10:28customer and try to execute for them. In
276. 10:31the early idea fog, where you can only
277. 10:34see 10 ft in front of you, the
278. 10:36temptation is to take a few cautious
279. 10:38steps in every direction. Sample a
280. 10:40little here, a little there, stay close
281. 10:43to home. The problem is that gives you
282. 10:45almost no information. What actually
283. 10:47works is to commit to one direction and
284. 10:50walk fast. You're not guaranteed to end
285. 10:53up in the right place, but you generate
286. 10:55much more information per unit of time.
287. 10:58And when you're walking, you might
288. 11:00arrive at a better destination,
289. 11:02one you couldn't have seen from the
290. 11:04start. The worst failure mode isn't
291. 11:07being wrong. It's not making a decision.
292. 11:10Spinning your wheels,
293. 11:12dabbling between ideas and never going
294. 11:14deep enough on any one of them to learn
295. 11:16anything. So, pick one and go deep.
296. 11:19Thanks for watching.
297. 11:21>> [music]
298. 11:27[music]