David (00:25)
If everybody vibe codes their own products, why is it that there are so few of them live in production?

Ilan (00:31)
Phil spent months trying to figure out how to use AI to create his own thing. And what he learned is that the biggest unlock is not technology.

David (00:52)
All right, so we are here today with Phil Mau Phil, what do people need to know about you?

Phil Mau (00:55)
Hi,

Great to be here. I'm a product manager at Early Trade for integrations. my history as engineer turned product manager. I went to UC Santa Barbara for computer science, and my first job was out of there at Procore. Procore IPO'd in 2021, and I was luckily part of the Rocket Ship Unicorn, where I started back in 2012 and ended up building out their developer platform and integrations. So anything you see on the API reference docs, the app marketplace, and any integrations that are there for

customers I had built.

David (01:26)
How did we get here? To explain this, we have to go back to June, where Phil was just chatting directly with Claude.

Phil Mau (01:32)
I think I was just kind of enjoying life and and whatnot and just you know doing my own thing. Like I'm a big golfer, pickleball, you know, it's summer in LA, so it's just yeah, I can play as much beach volleyball as I want. And I was just focused on other things. And then I think the job search was kind of the aha moment. You know, I'm seeing a lot of people just progressing so much faster in in in producing apps. They're getting consultancies where they're building apps for for people and they're just getting projects. So, you know, they're finding ways to to better their.

lives, making more money, getting jobs and stuff like that. And so I I just felt like I was behind and you know, I didn't want to be left behind and that's kinda when I made the pivot.

I did interviews since February when I entered into the job market job search market. I was looking for a job. you know, AI changed some things, I got two final stages and then, you know, either got passed up because of something like AI related, like we've completely

you know, overhauled due to AI and our strategy has shifted. So this role has basically left and, So I think that was, you know, after a couple of rejections, I was like, all right, I got to get into Claude Code.

David (02:34)
So, so when you started off, things were kind of bumpy and you were getting kind of stuck, and maybe lost in, all of the AI stuff Walk us through that.

Phil Mau (02:43)
I I kinda jumped into the deep end, right? Like right into the fire, just like when you started your job. I didn't really know what to expect. I I kinda knew I have a bunch of product manager friends, they have a chat, they've been building for the last like six months and I kinda felt behind. meanwhile I'm trying to look for a job. So I decided like, you know, to get a job I need to jump into Claude Code and figure it out.

I love cooking David knows I love making cookies I'm probably at 150 plus versions and so I was like all right like let's make a GitHub for recipes

I don't like using Google Docs to track it. I was using Claude Code or sorry, Claude Chat, for the longest time and I was like, Hey, like what happened here? And as we know

Early stuff, it can kind of read the context. The middle kind of gets a little hairy, it just forgets. And so I was like, all right, I need to build something. I need to get off of this. Even if this is built for me and nobody else uses it, like let's do it. So

I kinda jumped into there and honestly it worked great initially 'cause you just kinda use natural language and

It kind of one-shotted my application, And that was great. But as I was building, it just kinda kept running too, you know, it built too much. It tried to be too perfect on the the the initial one-shot. And I ended up just like fixing things and I never got to really see the application grow from the initial build. And that was very frustrating,

so it's like one step forward, but then like 10 steps back. So it just wasn't scaling. I didn't really have a great way to manage everything. Everything was just in my head or in the chat.

David (04:06)
so Phil, you were just, chatting with Claude chat and then, putting that context into Claude code and things were getting a little bit messy. and so you said some friends recommended, some different tooling for you.

Phil Mau (04:17)
I was able to kind of spec everything out. But now I had a bunch of stuff that was just in Claude chat in in the context, right? I didn't really have anywhere to put it. I could extract it to Google Docs. And that's when my product friend said, Hey, like, you know, why don't you use linear? 'Cause I was trying to figure out, you know, are people still using Jira? W what's going on in this day and age? And he said linear is probably where I should put my issue tracking and so on. So when I looked into it, I decided, hey, like, you know, LinkedIn Pl premium gave me three months for you to notion.

So I was like, all let's do Notion. I wanna I wanna do this right and I wanna take everything that I just did for two hours and put it into a process. So Notion kept all my you know product thinking, the the the high-level specs, the understanding of the roadmap, while linear had all the issues. And so once I talked to Claude Code or Claude Chat, I was able to develop a workflow and communication layer that used Linear and the session prompt in code to basically kick off the coding session. It would basically go read the linear ticket.

It would set up pre-flight, look up Superbase in Vercelle as it's hooked up and make sure things were validated and things actually existed instead of hallucinating that you know this session could actually st kick off. And so from that, I kind of used Linear to create the communication layer back and forth that instead of me having to take the prompt from Claude Chat and you know bring it to Claude Code and then bring the response back over, linear was my communication layer. It would post comments and and you know talk back and forth. So I could have my conversation in Claude Chat.

chat about what was going on when things got finished and use cowork to do QA on it instead of having to ask Claude Code who developed it, you know, if they did it right. And you know, this is also out of the need that, you know, GitHub MCP was broken on the chat side and so it had no way to look at the decisions that we had made and put into the docs decisions repo. we couldn't see like, you know, why we decided to do things, what actually got uploaded. So linear was our workaround and it actually became like the actual correct solution.

David (06:11)
And what was it like getting Claude Code connected to all these tools?

Phil Mau (06:15)
Simple. I mean the Supabase is all just like the API tokens throw into

y in the environment, linear, the same way, like everything just kinda connected either through the MCP plugins library or just through a token. you know, I obviously put guardrails in, like no production, only read, that kind of stuff. But you know, it really helped with the process and loading up because I think the biggest thing people will find is that with the hallucination, I think the configs th thought it was loaded or certain, you know, context was loaded, but it really wasn't. So the session command was kind of the game changer.

Ilan (06:47)
But there's something else to unlock.

The right tooling is vital, but there's something else that separates people who are just vibe coding a tool once versus people who are transforming how they operate. And it wasn't till a long drive that Phil discovered a new way to think about working with AI.

Phil Mau (07:05)
AI can lie to you. We all know it hallucinates. It'll you know, it'll say it built something and it just didn't. So the mindset shift is going from this kind of just, you know, narrative of just let's just

talk, right? Let's just talk it out. Let's build it as we go together, you know, hand in hand, to more of a structured business like we're

traditionally used to and so setting up workflows and understanding that you need to spec things out you know it'd be good to kind of have milestones or initiatives and stuff like that. You don't need to go too crazy because that's still running off like a Kanban board basically but just having better specs to give you know AI enough guardrails we'll say and give them a little autonomy to perform and really just getting me out of the loop at every step to kind of help shepherd the entire process. Now it's kind of you know I'm still there but I'm really there for product decision.

And I think that's kind of made the you know the process more enjoyable. I've seen more progress, it's less of fixing bugs and and kind of maintenance of infrastructure. I'm actually seeing things that I can go and and click around. So that's kind of really what's changed the way I use it on top of the tooling.

David (08:08)
You see, it's not just a great set of tools. You need a different way of thinking about working with AI. Build your app like a business and treat AI like a team, not just a bot to chat with.

Ilan (08:20)
what did you what did you build? I'm curious to learn more.

Phil Mau (08:23)
So Shikomi, I believe it means prep in Japanese. the normal stuff, like recipe something, were all taken, obviously, and people are sitting on domain.

So incorporated panda and corgis because those are my favorite animals.

Here's the recipe, basically the the recipe drop-down or the the list page. Everything was built kind of just from the initial one-shot. So there's definitely changes that I have made. I have a full Fable 5 you know redesign incoming. but for here, you can kind of see we got the nice collapse and so on. I'll go to my chocolate chip cookie recipe, and it's just a standard recipe book, you know, recipe page. It's it's kind of built a little looking technical because

I think I injected too much of GitHub in the prompt and you know traditional books, the old recipe books were kind of just very scientific. And so it kind of has this, you're able to do this. I incorporated things such as with Jules Chef steps like the Sous-Vide machine. There's a cook mode. So you can kind of cook and go through the steps and so on. So there's cool things like that. another view I can show you. You can, you know, my friend told me when I showed him like, hey, like it's always so frustrating when you're like, hey, add flour, and you're like, how much flour? Because so this cook mode.

eventually these are actually clickable so as you you know use the ingredients you can mark it as used and so that was like the biggest thing so I already got you know I guess customer demand and feedback so it was really cool to kind of see the the quick agile turnaround time

David (09:48)
you, hinted a little bit at it, but in terms of like talking to the market and, validating that it sounds like you had a good

Phil Mau (09:54)
Yeah yeah.

David (09:55)
time there. Yep.

Phil Mau (09:56)
Yeah, so you know, I had time validating it because obviously, you know, you built something impressive to you, something that's near and dear and and you know, I wanna show everybody. so hit up friends, hit up you know, people I knew who like to cook and already got, you know, feedback on, you know, I want XYZ and and the one was the cook mode that I'll show in a second. But you know, the friend asking me, Hey, like I wanna be able to know like this step says add flour, like how much flour. So this one is the the best demo of it and I built it, you know, right after

we chatted we have timers to like you know how long you need to s knead the dough when you go to certain steps you can you know on mobile you can check off that you've used the ingredients so it just kind of was nice to see a validation on my app

the biggest one obviously the the core piece of this is the history or the version control so it's gonna look you know somewhat like GitHub it's not gonna be as technical but you can see right here you can see the history on the left side so v7 to v1 you can see like what has changed and

And what was removed, what was modified, what was added. So that's kind of like the, you know, imagine I had 150 of these, you know, how would I do it? And eventually there's gonna be the the Sous chef AI that I wanna build, which will have like duplicate detection. So if I've already done it before, it's gonna tell me that I've already done it. And I have like notes fields where you can understand what happened with that cook. And I think that's kind of just like, you know, that's the app and that's the strength of it for someone who's just a home cook who wants to eventually launch a micro bakery.

to sell the cookies out of their house. Like this is something that, you know, I think I need.

Ilan (11:31)
actually is it live, by the way, that this application? Can anyone go and like sign up and start using it?

Phil Mau (11:34)
Yeah, yeah, yeah. I

mean any anybody can go look at it and you know I'll show you I'll show you one more thing because I think it's worth it. So, you know, I think this is live. You can go in, you can start free and you can sign up and whatnot. Google Auth is hooked up and so on.

Ilan (11:48)
Do you think that you actually had an advantage starting only, let's say three months ago, versus those other people? Like maybe they they might be doing things an

Phil Mau (11:58)
Yes.

Ilan (11:59)
older way and you got to like catch up to the current way of doing things?

Phil Mau (12:03)
Yeah, I mean I think that pairing with the the mindset of, you know

they're much more of kind of the traditional MBA or just kind of business minded and I'm much more coming from the software engineered turn product. So I think that paired with yes, like the tools just have completely evolved. Like you said, it's been like six months of honestly when claude code or just code has ramped up. Before I think it was, you know, still a little messy, still a little hard to use and you know, with Claude design coming out and and and being better over iterations, like I kinda had everything at my fingertips, which they didn't have. And so, you know

I'm not gonna say I built the a clone of my friend's app, but there's another app we built. Like I basically just kind of did that in the course of I think a month, and he's been working on it for six months. But that's also again, like he came in late, he also has a different kind of you know mindset and and background. And so it was kind of cool to see the comparison between you know what me and the other PMs are doing, and also the non-technical guy who's in the chat as well. Like everybody's using Claude Code and everybody's just using it a little different.

David (13:07)
All right, Phil. Thanks so much for walking us through that journey. Very interesting. Lots of things that people can learn there. Where can people find you?

Phil Mau (13:14)
Yeah, I mean you can find me at LinkedIn at at Malphilip, or you can find me at my site, get shicomi.com. it'll be linked probably in the comments. So hope to, you know, you guys enjoy the product.

Ilan (13:27)
Awesome.

Thank you so much, Phil, for for being on and and thank you for walking us through that journey.

Phil Mau (13:32)
Yeah, thank you.