Give Me 20 Minutes, I'll Make You AI Native
By Peter Yang
Summary
Topics Covered
- Voice Dictation Unlocks AI Speed
- Projects Capture 80% AI Value
- Prototype Before Specs Flips Waterfall
- Code Easier Than Documents Now
- Apps Disappear for Personal Agents
Full Transcript
Hey everyone. So today I want to show you how to become AI native in five steps. You know, everyone throws around
steps. You know, everyone throws around the terms AI native and AI PM without explaining how to get there. There's
like so much hype and nonsense out there. You don't have to read 100 plus
there. You don't have to read 100 plus articles or take $1,000 plus courses to become AI native. I went from AI novice to native in a year while balancing a
full-time job, my creative work, and kids. And you can do it much faster with
kids. And you can do it much faster with the right approach. So let's dive into it. Basically, I think there are five
it. Basically, I think there are five levels to become AI native. AI for
everyday answers, for daily work, for prototyping, for building apps, and AI as personal agent. So let's start with level one, AI for everyday answers. You
know, this level is where 99% of people are today. You're using chat GPT to get
are today. You're using chat GPT to get everyday answers instead of Google search. For example, the other day I was
search. For example, the other day I was able to fix my toilet by sharing a few questions and photos with ChibiT, which impressed my wife. This is where we all start, and that's fine. But if this is
all you're doing, well, you're barely scratching the surface. You should at least get to level two, which is AI for daily work. So, this level is actually a
daily work. So, this level is actually a massive step up from level one. And to
unlock level two, you should do three things.
First, you should use a voice dictation app like Whisper Flow, Monologue, Super Whisper, or something similar. It's just
so much faster to speak than to type.
And AI is excellent at understanding your stream of conscious thinking. Let's
give it a shot right now. So, I got Whisper Flow installed. And the great thing about Whisper Flow is I can just press a hotkey and voice dictate into any kind of text input box. I'm just
dictate for breakfast today, I'm going to have two eggs. uh yogurt,
blueberries, and my protein powder. And
you can see here that it's not only transcribed it, it's also made a list and formatted in a nice way. It's just
so much faster. Okay, so if you're not doing voice dictation, you should definitely do that starting today.
Second, you should use a meeting notes app like Granola. You know, having AI take notes has been a massive timesaver for every single type of meeting. and
Granola. They are a sponsor of this podcast, but they are a sponsor because they're, in my opinion, the best meeting notes app out there. So, the whole time I've been talking, I've been using
Granola to take notes here, right? At
any point, I can just ask during a meeting, what have people been talking about? And Granola can give me the
about? And Granola can give me the answer right here for me to catch up quickly. So, you can see how incredibly
quickly. So, you can see how incredibly useful this is for interviews, product reviews, any type of meeting. And unlike
other AI notes apps, Granola doesn't add annoying meeting bots to your calendar and so on and so forth. All right, I don't want to make this whole video a pitch for Granola, but I really do love the product. But the point is, use AI to
the product. But the point is, use AI to take meeting notes. If you're a PM, if you work in tech, probably have a ton of meetings, and this saves you just a ton of time.
Finally, you should have most of your conversations in AI projects. You know,
I have a dozen projects set up for product work, interviews, newsletter editing, and even weekend activity planning with my family. So, let me actually show you that project right
now. So, I have a cloud project for
now. So, I have a cloud project for weekend planning, and I giving it a bunch of context about myself, my family, and what we like to do. So, I'm
just going to ask it, um, what should we do today? Don't share any confidential
do today? Don't share any confidential information. All right. So, I'm going to
information. All right. So, I'm going to ask this. And what it's going to do is
ask this. And what it's going to do is it's going to look through a bunch of uh Bay Area parenting sites and hopefully come back with a recommendation. So,
let's just wait a little bit for this thing to generate. Okay. So, here we are. So, I'm recording this video on the
are. So, I'm recording this video on the morning of Super Bowl Sunday. And what
this thing has done is it searched through a whole bunch of uh family-friendly sites. As you can see
family-friendly sites. As you can see here, it searched through kind of local towns, what they list, and local events.
And let's see what it found for me. Let
me close this. So, it found a bunch of really awesome, really timely events from the aviation museum nearby to Lunar New Year celebration, two free family
days in San Francisco, and magnificent magnolia boom. There's just like a ton
magnolia boom. There's just like a ton of events that I can take my family to today. And I use this project all the
today. And I use this project all the time when I'm looking for activities to do with my kids on the weekend. So I
highly recommend that you set this up for yourself uh and your family to just have enjoyable weekends. All right, so projects are extremely useful and to be
honest, probably 80% of the value that I get from AI still comes from this level two. I even helped my 60 plus year old
two. I even helped my 60 plus year old parents set up voice dictation, set up meeting notes, and set up projects so they can use AI to monitor health and
plan their vacations better. Now, let's
move on to level three, which is AI for prototyping.
And this level fundamentally changed how I work as a product manager. The fact of the matter is it's now easier and way more fun to iterate on a prototype than
a document or design. And here is how I build products now instead of following a broken waterfall process where you have to do the spec and handle the spec to designer and designer hands out the
design and the spec to the engineers doesn't work anymore. Okay. So instead
what I do is I start with a prototype.
First I create a prototype using Google AI studio or replet by cloning my product's existing UI and adding a feature on top. Second I gather feedback fast. I should a prototype with my team,
fast. I should a prototype with my team, with leaders, with stakeholders to gather feedback and iterate quickly. And
you can even shoot a prototype with real customers, of course. And finally, after I feel confident in the prototype's direction, then I create a spec and design to iron out the edge cases and
more. Okay, so it kind of flips the
more. Okay, so it kind of flips the waterfall development process on its head. So let me quickly show you how
head. So let me quickly show you how this works.
Okay, so here I've used Replet to reproduce the YouTube studio UI of creators uh as a prototype, right? And
all I did was I pasted a screenshot of my YouTube studio dashboard and I just said make this for me and make it interactive. Keep the styling and
interactive. Keep the styling and everything very similar. So this is the actual YouTube studio. And now why don't we prototype a new feature? Let's
replace this news thing and what's new with something more useful for us which is content ideas for the video that I want to make next. So let's just say can
you replace news and what's new on the right with top five similar videos that my audience like to watch and an AI generated suggestion of what video I
should make next. So I'm replace it with a panel that's styled the same. Okay, so
you can see here that I basically just voice dictated again going back to step two my instructions and let's see what AI comes up with. Okay, so that took about 20 seconds and now it's
refreshing. And you can see here that it
refreshing. And you can see here that it did it. It replaced the right panel
did it. It replaced the right panel information with videos my audience likes to watch and it also had an AI generated content suggestion with a link to create the video. It's just way
faster to prototype again than to try to write the spec around this or to make a Figma design. If you're still stuck in a
Figma design. If you're still stuck in a team where people are chasing waterfalls, just start creating and sharing prototypes in meetings. I've yet
to meet a leader who doesn't prefer playing with a prototype over looking at a static doc or design. It just brings to life what you're trying to build so much better. And the bottom line here is
much better. And the bottom line here is that code is now easier to create than any kind of document or slide deck or design or intermediate artifact, right?
So definitely start prototyping to help you seek to truth much faster and just make your job way more fun. Okay, so now let's move on to the next level. So
level four is AI for building apps. And
this level is where you can't help but think about building apps with AI all day. Just 6 months ago, Vibe coding was
day. Just 6 months ago, Vibe coding was a painful process of trial and error.
Now I find it more fun than even watching Netflix. And I've built over 14
watching Netflix. And I've built over 14 apps to date. So let me show you some of them. This is what I'm working on right
them. This is what I'm working on right now. A builder trail app is similar to
now. A builder trail app is similar to Oregon Trail except you are a big tech PM who's trying to become a builder, right? And you can enter your name here
right? And you can enter your name here and pick a avatar and you can pick your favorite tool. Let's say clock code. And
favorite tool. Let's say clock code. And
then you can start shipping stuff and it will give you like scenarios for you to do and your goal is to ship five apps before you run out of energy or momentum
or savings. So, you can just go through
or savings. So, you can just go through this and ship it and keep going. And I'm
still working on the game mechanics, trying to make it more fun, but you can see that this is just a thrill to build.
All right, let me show you another app that I built. So, this one is just a personal website that I built. Pretty
straightforward. It's at petergen.com if you want to check it out. And I've
integrated with Substack, so it pulls my most recent post here. And down here there's an Easter egg where I actually highlight my favorite movies and
favorite games that I played growing up.
Chrono Trigger, Starcraft, RU War, Attack on Titan, amazing anime, MadMax, you get the idea. Yeah, it's super easy to spin this up. I use uh Nano Banana to
generate this image and I think the site looks great. So definitely recommend
looks great. So definitely recommend that you build a personal website too with some Easter eggs. All right. So,
next, uh, let's go to board game, actually. So, I was playing board games
actually. So, I was playing board games with my kids. We love doing it. I just
want to find board games that we actually love playing. So, I can filter here by minimum ratings, by play time, and so on so forth. And just to kind of
find a game that we actually like to play. And the way I built this was I
play. And the way I built this was I used Replet. I uh got data from
used Replet. I uh got data from boardamegeeek. If you go to Kaggle,
boardamegeeek. If you go to Kaggle, there's like a bunch of uh board game information. and it's just like a much
information. and it's just like a much better interface in my opinion to browse these board games than the actual default board game geek interface. So,
I'll include links to all these sites and the stuff that I'm showing here in the video description.
Okay, and last but not least, I built this retro space shooter. So, let's
check it out. So, yeah, here we go. It's
got awesome graphics uh because I uploaded some pixel art and I just launch it. And there you go. So you it's
launch it. And there you go. So you it's a fully playable space shooter with enemies that you can destroy, boss battles and so on. Probably took me I don't know two hours to make this and it
that was like while watching Netflix and actually I have a full video tutorial about this that also link in the description. But yeah, this is just
description. But yeah, this is just super fun. This kind of brings me back
super fun. This kind of brings me back the old games I used to play as a kid and you can build something very similar too. Okay. All right. Let let me close
too. Okay. All right. Let let me close close this. All right. So, at this
close this. All right. So, at this point, you might be thinking, "Look, Peter, this sounds really cool, but I really have no idea how to code or what a terminal even is. Someone needs to
show me how to do this." Well, I think that's why you're watching these videos.
But, I also want to remind you that at the end of the day, you're just chatting with the AI agent using plain English.
If you're just vioding these games and personal apps, you really don't have to look at the code at all. And as much as I love making tutorials, I do think AI is a better teacher than even me. And a
terminal, if you get intimidated by it, it's just another interface to chat with the AI agent. Think of it as just another chat window.
Okay, so before we move on from this step, let me try to answer two questions. First, how can you start
questions. First, how can you start building and what should you build?
Right? So I think these three tips will take you a long way. So, if you don't know what to build, just tell AI your problems. Ask it, "What are some simple apps that we can build to help me take
time back? And here's a bunch of context
time back? And here's a bunch of context about my life." Or just build something fun like a game just to start. Number
two, create a robust plan first. Ask it
to create a plan with three clear milestones to avoid mistakes during coding. The reality of actually building
coding. The reality of actually building with AI and coding with AI is that you should actually spend most of your time planning and just writing English documents. and the AI does the actual
documents. and the AI does the actual coding for you, right? So definitely
make up time to plan well up front. And
finally, think of yourself as the manager and AI is your very capable intern who's doing all the work. Give it
feedback, share screenshots of bugs with it, give it direction just like you would a human employee. Be patient with it and don't give up and I promise you'll build something amazing. All
right, so next question is, which AI coding tools should I use? There are so many tools out there. I think just start building instead of worrying about which tool to use. But if you want my opinion, here's my rough guide. If you're just
getting started, I would use Replet. It
is a full stack platform with database authentication, logging, and most importantly, security builtin. I've seen
people vibe stuff with some of these other web app builders and run into much of a security issues and API keys leaking. Replet protects you against all
leaking. Replet protects you against all that, right? So personally I still use
that, right? So personally I still use Repetit to prototype and build websites.
Now if you want more control give Cursor a try. It's the most popular IDE for a
a try. It's the most popular IDE for a reason. But to actually deploy with
reason. But to actually deploy with Cursor you have to know how to set up tools like GitHub and Verscell to host your apps. And it's not too complicated.
your apps. And it's not too complicated.
You just have to ask the AI to deploy your apps for you. Now if you want maximum power, consider using clock code in a terminal or codeex app which OpenAI
recently made available. Right now my preferred tools are a terminal app called Ghosty and Clockode and I also use this thing called Happy Engineering
to interact with it on my mobile phone.
So again I'll provide all the links in the video description. All right. So the
best engineers like Peter who made OpenClaw are using apps like Conductor or multiple terminal windows to run a bunch of AI agents at once. This is this
really funny meme that someone shared where you start by like coding in a web interface and accepting all of AI's suggestions, right? And then you go to
suggestions, right? And then you go to YOLO mode which is AI can just make changes without you having to prove it at all. And at some point you go to
at all. And at some point you go to clock code or codecs and at some point you start having multiple clock code terminal windows so you can build multiple features for the product at the same time. And eventually you get to
same time. And eventually you get to figure eight, which is you're using one AI to orchestrate a bunch of other AI agents. Now, this sounds really
agents. Now, this sounds really complicated. I think even Peter is
complicated. I think even Peter is probably still at figure 7. I'm not sure how many terminal windows he has open here. He's probably still at figure 7,
here. He's probably still at figure 7, but uh don't get intimidated by all this stuff. Just start coding. It's like
stuff. Just start coding. It's like
super fun. It's arguably the most fun video game that you can play. Okay, now
let's talk about level five. Using AI as your personal agent. You know, it's easier than ever to build an app with AI these days, but I think most apps will actually disappear. Instead, we'll all
actually disappear. Instead, we'll all be spending our time onboarding and giving tools to our agents like my conversation with my open claw bot here.
And I think the UX that we know today, buttons, menus, navigation will all go away. The ultimate UX is lying in bed
away. The ultimate UX is lying in bed just texting my personal AI agent friend, giving it vague directions via voice or text and having your agent just
get you and get stuff done for you. Let
me kind of show you my open cloud bot again. So here's my open clubbot and let
again. So here's my open clubbot and let me just say without sharing any competition information, tell me what you can do. Okay, so it's come up with a list of things you can do and this is
after spending a lot of time setting it up and onboarding it, right? just like I would an employee. But right now, it can do web searches. It can summarize articles for me. Can read and write
files. It can do a bunch of other stuff.
files. It can do a bunch of other stuff.
So, let's just ask it, what are some things that you're helping me do these days, right? Let's see what it comes up
days, right? Let's see what it comes up with. Okay. So, you can see here that
with. Okay. So, you can see here that I've set it up to triage my emails, to manage my calendar, do research, help me on my writing, task reminders. I think
it's not getting into specifics because I told you not to share any confidential information. But needless to say, uh
information. But needless to say, uh this thing now has read access to my banking information. It has my YouTube.
banking information. It has my YouTube.
It can edit documents for me. I made a lot of OpenCloud tutorials, so you can check those out. But I'm slowly onboarding my OpenCloud bot, Zoey, to be more and more useful to me while trying
to keep it as safe as possible. Okay.
Now, this level is not just about OpenCloud, right? And open client in
OpenCloud, right? And open client in some ways is a very janky and early version of what a personal AI agent can look like. So to make this more
look like. So to make this more concrete, let me talk about three people who I think are at level five. So the
first one is Kieran who is an entrepreneur in residence at every and Kieran has built a full clock system to make it smarter each time he uses it. He
calls it the compound engineering system. So he's not just using claw
system. So he's not just using claw code, he's fully customized claw code for his needs. And this system has a set of custom skills that he's built and includes this command called LFG command
that he set up where once you run this command, it does all these steps for you. So he has a bunch of other skills
you. So he has a bunch of other skills for planning, for working for review and doing all this stuff and even making like a demo video for his new feature that he set up. So all he has to do is
just run this LFG command which by the way means let's f and go and then you can go away for two or three hours and come back hopefully to a fully built product. All right so if you use clock
product. All right so if you use clock code a lot you're inevitably end up customizing it with your own commands and Kieran in my opinion has taken it to the next level and you can find his LFG
command his compound engineering system in the link in the description as well.
The second person I want to show is Teresa, who uses Claw Code and Obsidian, a note-taking app to run her life and business. She has built three layers of
business. She has built three layers of context files that lets Claude generate her daily to-do list, draft blog posts, and help her manage her tasks, right?
So, what she's done here is basically she has a note-taking app here that has all her stuff, and she has two clock terminals that she uses to update her Obsidian note-taking fault. And there's
like a lot of notes here, right? But a
lot of this stuff is actually AI generated by her just talking to cloud code in her terminal windows. Now,
personally, I love to try this myself, but most of my context is in Google Drive, which is why I still rely on OpenClaw and some of these other hacks.
Um, but yeah, you can definitely give Claw Code and Obsidian a try yourself to set up a personal agent. And finally, my friend Nat Elias, okay, he went to the
next level. He set up an open cloud bot
next level. He set up an open cloud bot to build a business autonomously. He
gave it its own X account and its own API keys to Stripe, GitHub, and more.
And amazingly, after just a week, his agent, which he calls Fetus Craft, has made $3,500 in revenue by selling a PDF for people to set up Open Claw in a
similar way. And you can see here he's
similar way. And you can see here he's also made $37,000 in trading fees from crypto because someone set up a Felix crypto token.
Now, I don't know about the crypto stuff, but the fact that he set up a website and sold 132 copies of this PDF largely by himself is kind of mind-blowing. And I have an awesome
mind-blowing. And I have an awesome interview that I'll share with Nat soon about exactly how this worked. Okay, but
basically I think the interface of the future is just lying in bed and asking your AI to do things for you in the most lazy way possible. And the three people above are already there. They're
messaging AI to do a variety of chores while they're living their life. Okay,
so to recap, here's how to get started to become AI native, right? So level
one, you're just using chat cloud or Gemini instead of Google search to ask everyday questions. Level two, you've
everyday questions. Level two, you've set up voice dictation with whisperflow.
You set up granola with meeting notes and you're using AI projects. Level
three, you are using prototyping in replet, Google AI studio or another tool. And when you're building products,
tool. And when you're building products, you're starting with a prototype instead of the spec or all this internal artifact crap that no customer will ever see. At level four, you are just
see. At level four, you are just thinking about building apps with AI all day. To get started, just build a game
day. To get started, just build a game or pick a problem that you have and vibe code an app to solve it. And finally,
level five is you are setting up a personal agent system that learns from every session you have with it that helps you run your life, run your work, and run your business. Now, the point is
you don't want to jump straight to level four and five, right? A lot of the value is actually still in the earlier levels.
For me, projects and prototyping have just saved me so much time at work, and they're just like incredibly easy to set up. So, as you navigate the levels shown
up. So, as you navigate the levels shown here, just remember to use AI to solve your own problems. Try things and just have it fun. It shouldn't feel like work. It should feel like the most fun
work. It should feel like the most fun video game that you've ever played. And
if you're focused, I think you can actually scale all the way up to level four in a week. Level five might take a little bit longer, but level four, I think you can do in a week. So again,
just start using AI and scale the levels yourself. All right, and of course, be
yourself. All right, and of course, be sure to like and subscribe to my channel for more practical, no hype AI tutorials and videos. We're all on this journey
and videos. We're all on this journey together. All right, so that's it for
together. All right, so that's it for today, and I'll see you at level five.
Loading video analysis...