How to Build an MVP in the AI Coding Era

1. 0:00 – 1:19

   Why MVP advice needs updating for the AI coding era

   1. MSMichael Seibel0:00

      I think for better or for worse, the consistent problem has always been how do you help your customers be more successful? If you had asked me five years ago, "Oh, the price of building features has decreased, that's obviously good," I think I would've probably said yes. And I think that, like, the TLDR of this video is, like-

   2. DCDalton Caldwell0:20

      Unclear.

   3. MSMichael Seibel0:23

      [upbeat music] This is Dalton + Michael, and today we're gonna talk about how to build an MVP in the age of AI. This is interesting topic, 'cause I actually think we have to disagree with our former advice.

   4. DCDalton Caldwell0:37

      Yep.

   5. MSMichael Seibel0:37

      Or there's a different set of constraints now. It used to be the case you had to figure out what's the smallest set of features to build that would create value, because building features was costly. We're now in maybe a different situation. [laughs]

   6. DCDalton Caldwell0:52

      [laughs]

   7. MSMichael Seibel0:53

      What, what environment are we in now? [laughs]

   8. DCDalton Caldwell0:55

      I think that's completely right. I think much has been said of how to build an MVP and get first customers, and there's-

   9. MSMichael Seibel1:00

      Yes

   10. DCDalton Caldwell1:00

       ... you know, we've talked about it. There's great content out there. But I think that advice needs updating-

   11. MSMichael Seibel1:04

       Yes

   12. DCDalton Caldwell1:05

       ... to AI tools.

   13. MSMichael Seibel1:06

       Yes.

   14. DCDalton Caldwell1:06

       And I could speak to this myself a little bit-

   15. MSMichael Seibel1:08

       Mm-hmm

   16. DCDalton Caldwell1:08

       ... because I recently built a product and, and launched it. You know, it's called StandardDB if people wanna check it out. But we won't, we won't talk about the specifics, but-

   17. MSMichael Seibel1:16

       Mm-hmm

   18. DCDalton Caldwell1:17

       ... guys, I've gone, I've gone through this. I just built and launched a product.

   19. MSMichael Seibel1:19

       Yes.

2. 1:19 – 2:20

   AI-driven feature creep: when building is too easy

   1. DCDalton Caldwell1:19

      And so let me start with what's different. Number one, it is so much easier to build features-

   2. MSMichael Seibel1:24

      [laughs]

   3. DCDalton Caldwell1:25

      ... that for me to launch that, I had to delete 80% of the features that I built in the MVP.

   4. MSMichael Seibel1:33

      That's, that-

   5. DCDalton Caldwell1:34

      [laughs]

   6. MSMichael Seibel1:34

      ... that is not an old problem. That is a very new problem. [laughs]

   7. DCDalton Caldwell1:37

      I built this crazy, huge thing with my co-founder Brian in two weeks, and I'm like, "Oh, this is bad. This is, this is, like, a bunch of crap."

   8. MSMichael Seibel1:47

      Gotta cut. [laughs]

   9. DCDalton Caldwell1:47

      And so I had to, I had to cut.

   10. MSMichael Seibel1:49

       Yeah.

   11. DCDalton Caldwell1:49

       And that feels very different to me-

   12. MSMichael Seibel1:51

       Yes

   13. DCDalton Caldwell1:51

       ... than just trying to build a product.

   14. MSMichael Seibel1:54

       Yeah.

   15. DCDalton Caldwell1:54

       Like, the difficulty of building something prevented you from, from getting into that feature creep.

   16. MSMichael Seibel1:57

       Yeah. Or that feature creep would take months.

   17. DCDalton Caldwell2:01

       Yes.

   18. MSMichael Seibel2:02

       Whereas you're basically, you got feature creep in two weeks.

   19. DCDalton Caldwell2:04

       I got feature creep in two weeks.

   20. MSMichael Seibel2:04

       [laughs]

   21. DCDalton Caldwell2:04

       And so you see people just vibe coding just crap.

   22. MSMichael Seibel2:08

       Yes.

   23. DCDalton Caldwell2:09

       And it's because it's so easy to just keep adding and adding and adding.

   24. MSMichael Seibel2:13

       Yes.

   25. DCDalton Caldwell2:14

       And so I think that you almost have to be more of an editor than you did in the past.

   26. MSMichael Seibel2:19

       That's a big one.

   27. DCDalton Caldwell2:19

       Right?

3. 2:20 – 2:59

   The temptation to avoid users and build “everything they’d want”

   1. MSMichael Seibel2:20

      I might argue it also makes it far more challenging when you're building a product for someone else. 'Cause when you're building a product for yourself, at minimum you can use it throughout the MVP process and see if you're getting value. When you're building a product for someone else and you can build infinite features, your brain can be like, "Well, they would want this feature."

   2. DCDalton Caldwell2:39

      Yeah.

   3. MSMichael Seibel2:39

      "They would want that feature. They would want..." Why would I wanna go to them with something that doesn't have all the features they're gonna want?

   4. DCDalton Caldwell2:44

      Right. Why don't I give them all the features they want right in the beginning? [laughs]

   5. MSMichael Seibel2:47

      Well, and it's addicting-

   6. DCDalton Caldwell2:48

      Yeah

   7. MSMichael Seibel2:49

      ... to write the feature. It's fun to, like, tell Codex.

   8. DCDalton Caldwell2:51

      [laughs] Yeah. Yeah.

   9. MSMichael Seibel2:52

      And so I think what I'm arguing is it's more tempting now than ever-

   10. DCDalton Caldwell2:57

       Yes

   11. MSMichael Seibel2:57

       ... to not talk to users-

   12. DCDalton Caldwell2:59

       Yes

4. 2:59 – 4:19

   Why building every requested feature is now dangerously possible

   1. MSMichael Seibel2:59

      ... and to not launch with something small. I wanna extend that. And when you talk to users and they gave you the laundry list of features, in the past you knew you couldn't build all that, so, or at least smart people did, and so your instinct would be, "Let me try to dig deeper. Let me try to figure out what they really need, what's really going on." Now when they hand you a list of features, like, "Hey, can you describe those in more detail?"

   2. DCDalton Caldwell3:21

      I mean-

   3. MSMichael Seibel3:21

      We can literally just give those to- [laughs]

   4. DCDalton Caldwell3:23

      Yeah. Let me make an extreme point. You could take notes-

   5. MSMichael Seibel3:26

      Yes

   6. DCDalton Caldwell3:26

      ... with Groena of you talking to a user and feed them to Codex, and Codex will build every feature the user asked for.

   7. MSMichael Seibel3:34

      Yes.

   8. DCDalton Caldwell3:34

      And I'm saying that's really bad. [laughs]

   9. MSMichael Seibel3:36

      Yeah. [laughs]

   10. DCDalton Caldwell3:37

       Like, don't do it.

   11. MSMichael Seibel3:38

       That's the counterintuitive bit, is it's always been a theorem build, to build every feature a user wants, but now it's an exponential theorem because you can. Like, you actually could do that, and that's really, really, really bad. Unfortunately, I think as a result, knowing yourself about the problem is way more important now. Because, like, if you don't know about the problem and users are telling you features and they're excited and you don't know what the hell's going on, you can get yourself in... Like, you, you, you basically have created a tar pit.

   12. DCDalton Caldwell4:11

       Yes.

   13. MSMichael Seibel4:12

       Like, like, like, where one never maybe have existed, you, you're, you are the tar pit. [laughs]

   14. DCDalton Caldwell4:16

       Yeah. And, and again, I'm saying I experience, like, I, I feel this viscerally-

   15. MSMichael Seibel4:19

       Yes

5. 4:19 – 4:59

   Resisting broad complexity: earn features through user pull

   1. DCDalton Caldwell4:19

      ... where it's so easy to crank stuff out. And so I think the advice here, founders, is you have to try harder than ever-

   2. MSMichael Seibel4:27

      Yes

   3. DCDalton Caldwell4:27

      ... to resist the temptation of building super complex, broad surface area products. And of course you should add more features to it, but it should be because users are pulling you-

   4. MSMichael Seibel4:39

      Mm-hmm

   5. DCDalton Caldwell4:39

      ... towards building more stuff, I think.

   6. MSMichael Seibel4:41

      Mm-hmm.

   7. DCDalton Caldwell4:41

      And not because Codex can crank it out. [laughs]

   8. MSMichael Seibel4:45

      Yeah. Well, let's talk about that users bit. So okay, it's easier to build products. I was about to say I can't believe I'm saying this, but of course I can believe I'm saying this. Spamming is easier, too. Spamming always gets easier. [laughs] Like, that's the history. [laughs] So let's talk about that.

6. 4:59 – 5:40

   “Talking to users” vs. AI-enabled spam: quality over quantity

   1. DCDalton Caldwell4:59

      You know, the old advice is have 100 people that love your product. The old advice is talk to as many users as possible. Again, it's all true.

   2. MSMichael Seibel5:06

      Yes.

   3. DCDalton Caldwell5:06

      We're not saying that's wrong. Here's the problem. It's easier than ever to have your claw spam people-

   4. MSMichael Seibel5:13

      Yeah

   5. DCDalton Caldwell5:13

      ... or you to go buy a bunch of leads online-

   6. MSMichael Seibel5:15

      Yes, yes

   7. DCDalton Caldwell5:16

      ... and spam people.

   8. MSMichael Seibel5:17

      Yes.

   9. DCDalton Caldwell5:17

      And so a lot of founders, their version of talking to users-

   10. MSMichael Seibel5:21

       It's perverted. [laughs]

   11. DCDalton Caldwell5:22

       ... is to, it's to, like, spam.

   12. MSMichael Seibel5:24

       Yes.

   13. DCDalton Caldwell5:24

       And it's easy, and the numbers are big.

   14. MSMichael Seibel5:26

       Yes.

   15. DCDalton Caldwell5:26

       And again, so here's the counterintuitive thing. Don't do it.

   16. MSMichael Seibel5:30

       Yeah.

   17. DCDalton Caldwell5:30

       [laughs] That is noise.

   18. MSMichael Seibel5:32

       Yes.

   19. DCDalton Caldwell5:33

       And a smaller number of high-quality conversations where you really listen-

   20. MSMichael Seibel5:38

       Yeah

   21. DCDalton Caldwell5:38

       ... is much better than spamming 1,000 people.

7. 5:40 – 6:19

   Users don’t have the full playbook—go deeper than feature requests

   1. MSMichael Seibel5:40

      If we wanna go one step deeper, right, if startups were as simple as talk to users, build what they want, so many more startups would be successful. Fundamentally, you have to assume that most of your users want their business or their project to be successful, and they don't exactly know how. So some fundamental part of what you're doing... is 10X harder because you're trying to figure out how to help your customer's business do better, and your customer doesn't have that playbook perfectly. By talking to a smaller number of customers more intimately, it massively increases the chances that you can actually make an impact in their business. And then that spins a flywheel that's not spam, it's word of mouth.

8. 6:19 – 7:37

   StandardDB lesson: fewer features made the product click

   1. DCDalton Caldwell6:19

      A- again, let me give you my real example for what I did with StandardDB.

   2. MSMichael Seibel6:22

      Yes.

   3. DCDalton Caldwell6:22

      We built a bunch of stuff, and then I set up 12 Zoom calls-

   4. MSMichael Seibel6:27

      Yes

   5. DCDalton Caldwell6:28

      ... with founders that I knew, that I- that are in the- that could be users of it.

   6. MSMichael Seibel6:32

      Yes.

   7. DCDalton Caldwell6:32

      My cofound- founder and I joined, and I- we just asked them. And a lot of people didn't really understand what it was-

   8. MSMichael Seibel6:39

      Yeah

   9. DCDalton Caldwell6:39

      ... because it was too complicated.

   10. MSMichael Seibel6:40

       Yeah.

   11. DCDalton Caldwell6:41

       And my... I tried to explain what it was, and like, "Oh, I don't know."

   12. MSMichael Seibel6:44

       [laughs]

   13. DCDalton Caldwell6:44

       And so we- when we stripped out 80% of the functionality-

   14. MSMichael Seibel6:48

       Yeah

   15. DCDalton Caldwell6:48

       ... they were like, "Yeah, I, w- I'm signing up right now."

   16. MSMichael Seibel6:51

       There you-

   17. DCDalton Caldwell6:51

       Like, someone signed up on the call.

   18. MSMichael Seibel6:52

       [laughs] There you go.

   19. DCDalton Caldwell6:53

       And so-

   20. MSMichael Seibel6:53

       That's counterintuitive. Less features, more usage. [laughs]

   21. DCDalton Caldwell6:56

       Isn't that weird?

   22. MSMichael Seibel6:57

       Yes.

   23. DCDalton Caldwell6:57

       And it's because they didn't- they were, like, overwhelmed-

   24. MSMichael Seibel6:59

       Yeah

   25. DCDalton Caldwell6:59

       ... by too much complexity.

   26. MSMichael Seibel7:01

       Yes.

   27. DCDalton Caldwell7:02

       And my vision, I had this expansive vision of what it could be, but man, like, users don't care about that. They just wanna know how this solves their short-term problems.

   28. MSMichael Seibel7:08

       No.

   29. DCDalton Caldwell7:09

       Right?

   30. MSMichael Seibel7:09

       No. I think for better or for worse, the consistent problem has always been, how do you help your customers be more successful? And if you had asked me five years ago, "Oh, the price of building features has decreased, that's obviously good," I think I would've probably said yes. And I think that, like, the TLDR of this video is, like-
9. 7:37 – 9:20

   GarageBand metaphor: democratized tools don’t increase demand or winners

   1. DCDalton Caldwell7:37

      I'll tell you about it. Think about how hard it was to get a record made in the '60s.

   2. MSMichael Seibel7:45

      Mm.

   3. DCDalton Caldwell7:46

      You needed a label deal, you had to pay for studio time, you needed instrumen- you know, it was, like, super expensive. How many people created a radio-quality record in the '60s? Thousands?

   4. MSMichael Seibel7:57

      Mm-hmm.

   5. DCDalton Caldwell7:59

      Now, you buy a Mac-

   6. MSMichael Seibel8:00

      [laughs]

   7. DCDalton Caldwell8:01

      ... and it comes with GarageBand.

   8. MSMichael Seibel8:03

      Yes.

   9. DCDalton Caldwell8:03

      And a lot of number one hits, you can research this, were made-

   10. MSMichael Seibel8:06

       Built there, yes

   11. DCDalton Caldwell8:06

       ... in GarageBand.

   12. MSMichael Seibel8:07

       Yes.

   13. DCDalton Caldwell8:07

       So you're using the same tools-

   14. MSMichael Seibel8:09

       Yes

   15. DCDalton Caldwell8:09

       ... that the greatest people in the world are using.

   16. MSMichael Seibel8:11

       Yes.

   17. DCDalton Caldwell8:11

       Now, how much more demand is there for music than there was in the '60s?

   18. MSMichael Seibel8:18

       Mm.

   19. DCDalton Caldwell8:18

       It's flat.

   20. MSMichael Seibel8:19

       Mm.

   21. DCDalton Caldwell8:19

       Like, how many hours a day can people listen to music?

   22. MSMichael Seibel8:22

       People are not living more than 24 hours a day. [laughs]

   23. DCDalton Caldwell8:24

       And how many people wake up every day being like, "Man, I really wanna hear some songs written by, you know, some guy in Illinois-

   24. MSMichael Seibel8:31

       That no one's ever heard of [laughs]

   25. DCDalton Caldwell8:31

       ... no one's ever heard of. Gee, I really..."

   26. MSMichael Seibel8:33

       [laughs]

   27. DCDalton Caldwell8:34

       Like, like, no one wants that.

   28. MSMichael Seibel8:36

       Yeah.

   29. DCDalton Caldwell8:36

       And so this is my metaphor for a lot of the vibe coding stuff-

   30. MSMichael Seibel8:39

       Yeah
10. 9:20 – 10:12

    Where AI tools may help most: empowered employees vs. startups

    1. MSMichael Seibel9:20

       Well, and what's funny about this is that when you're building tools for yourself, um, you know, I would argue that, like, an employee inside of a company is now so much more empowered with these tools. When they have a clear job to do and they know how to do it-

    2. DCDalton Caldwell9:33

       Yep

    3. MSMichael Seibel9:33

       ... and now it's like they don't have to tax the company's resources, they probably get a much higher multiplier effect than the startup.

    4. DCDalton Caldwell9:41

       It's true. I mean, it's the, it's like think about how many people are employed, um, making videos for, like, internal training purposes or for whatever-

    5. MSMichael Seibel9:50

       Yes

    6. DCDalton Caldwell9:50

       ... versus people that are making Hollywood blockbusters.

    7. MSMichael Seibel9:52

       [laughs] Yes, yes.

    8. DCDalton Caldwell9:52

       You know? Like, or how many directors are in the world.

    9. MSMichael Seibel9:54

       Yes, yes.

    10. DCDalton Caldwell9:54

        And so probably there's lots and lots of people who use these tools to get their job done-

    11. MSMichael Seibel9:58

        Yes

    12. DCDalton Caldwell9:58

        ... that might not be creating the next Facebook.

    13. MSMichael Seibel10:00

        Exactly, exactly.

    14. DCDalton Caldwell10:01

        [laughs]

    15. MSMichael Seibel10:02

        So TLDR, building an MVP in the AI world, different. Features, I've always thought features are bad, but now features are cheap. Oof, scary.

11. 10:12 – 11:03

    Build-in-public in the AI era: avoid content slop, be idiosyncratic

    1. DCDalton Caldwell10:12

       You just have to be picky, and, like, maybe one more thing is a lot of our advice was always to build in public and to publish a lot about what you're doing.

    2. MSMichael Seibel10:20

       Mm-hmm.

    3. DCDalton Caldwell10:20

       And the problem is the claws make it so easy-

    4. MSMichael Seibel10:24

       Yeah

    5. DCDalton Caldwell10:25

       ... to create crap on LinkedIn and X about what you're building, and it's all the same.

    6. MSMichael Seibel10:31

       Yes.

    7. DCDalton Caldwell10:31

       Versus if you take the time to write really unique stuff, you can build a following. And so I j- again, I see a lot of founders just putting out slop versus, like, Arkle, he writes about file systems all the time, Hunter does constantly.

    8. MSMichael Seibel10:43

       Yes.

    9. DCDalton Caldwell10:44

       And he's building a following 'cause it's super idiosyncratic writing about file systems.

    10. MSMichael Seibel10:48

        Yes.

    11. DCDalton Caldwell10:49

        Ironically also, the best content to learn about what the labs are working on is the Twitter accounts of people that work at the labs.

    12. MSMichael Seibel10:57

        Mm-hmm.

    13. DCDalton Caldwell10:57

        And they are definitely not writing all that stuff [laughs] for the LLMs.

    14. MSMichael Seibel11:00

        [laughs]

    15. DCDalton Caldwell11:01

        Do you get what I'm saying?

    16. MSMichael Seibel11:02

        It's a weird-

    17. DCDalton Caldwell11:02

        Like, they're leaking alpha.

12. 11:03 – 12:21

    Fake traction loops and the competitive edge of focus and editing

    1. MSMichael Seibel11:03

       Exactly. It's a weird game. It's basically like it's become easier for a startup to basically look like it's going through a positive feedback loop when it's actually secretly going through a negative feedback loop.

    2. DCDalton Caldwell11:14

       Yep.

    3. MSMichael Seibel11:14

       Spam a bunch of customers, get their usage, they don't like the product, they all churn out, keep doing it, da da da da. But that's an advantage in a weird way. Like, if more of your competition's doing that-

    4. DCDalton Caldwell11:25

       Great

    5. MSMichael Seibel11:26

       ... and you're not, you can figure out how to use these tools in the measured way to actually create value for people. There's tons of advantage to pick up right now. Tons of advantage.

    6. DCDalton Caldwell11:37

       And it's almost like the people that are the best at editing or creating white space-

    7. MSMichael Seibel11:44

       Yeah

    8. DCDalton Caldwell11:44

       ... and focusing-

    9. MSMichael Seibel11:45

       Yes

    10. DCDalton Caldwell11:46

        ... I think are, like, really well-positioned when everyone else is doing the opposite.

    11. MSMichael Seibel11:51

        Yes.

    12. DCDalton Caldwell11:51

        Like, that's a way to stand out from the crowd.

    13. MSMichael Seibel11:52

        Yes.

    14. DCDalton Caldwell11:53

        Right?

    15. MSMichael Seibel11:53

        We have two features that work-

    16. DCDalton Caldwell11:55

        [laughs]

    17. MSMichael Seibel11:56

        [laughs] ... that actually make your business better, uh-

    18. DCDalton Caldwell11:58

        Yeah

    19. MSMichael Seibel11:58

        ... or make your life better, yes, as opposed to 150 that you might be confused initially. Obviously, the thing with 150 features is better. Then when you start using it, you're like, "No, thank you."

    20. DCDalton Caldwell12:08

        You got it.

    21. MSMichael Seibel12:09

        All right. Cheers, Dalton.

    22. DCDalton Caldwell12:10

        All right, thanks. [laid-back music]

Episode duration: 12:22