Alex

Building a personal AI health coach that knows you and your goals @getOptima

[Alex @itsAlexGuerrero]

I lost 100 pounds.

Finally had complete control over my physique.

Bulk, cut, maintain - whatever I wanted.

I thought I beat the "health and fitness" game.

But I was wrong...

[Alex @itsAlexGuerrero]

I let AI write 99.9% of my code for my mobile app over the past month. Here's 23 things you need to know:

1. Always create a markdown file for the feature implementation. If you're not keeping track of where you are in the process (and more importantly the LLM isn't keeping track) then you will get lost in the weeds or go down rabbit holes. Tell the LLM what you want the feature to look like and go back and forth with it until you agree on what must be done. If you are unsure just ask it. The more detailed the plan the better. Also be sure to update the plan as you progress with the current status of the feature.

2. ALWAYS review the plan the LLM provides you. It's tempting to just blindly trust it but sometimes it will create unnecessary complexity that you will end up having to clean up later. Question everything it's doing and if it's absolutely necessary. Look into all the details.

3. If you start building out a feature and then realize it's not going the way you planned or you want to completely change the approach - it's okay, change it. Wasted time is better than an entire feature you don't like or doesn't work as you expect.

4. If you're using 3rd party services/integrations - make sure you tell the LLM to first review the official documentation from the provider before planning out any functionality. Sometimes it will hallucinate what will work and if you tell it to reference the official docs it will save you a ton of headache.

5. Tell the LLM to double check it's work. Sometimes it will hallucinate stuff and/or just do things you don't want it to do. Often times it will catch things it didn't on the first pass through.

6. Know what you're building for. This one seems pretty obvious but you need to put some thought into it before you begin. Web app, mobile app, desktop app, chrome extension etc. Once you know what you're building for identify the best tech stack for your project - again ask the LLM (or multiple LLMs) to get this answer.

7. When implementing a feature or making changes tell the LLM to approach it in a phased manner with a pause after each phase. Do this so you can manually verify that all things are working as expected and there's not a bunch of tech debt being added to the project. You can also tell it to create tests to make sure it's working but I like to manually verify first so I know for sure that it's good.

8. If you are dealing with a bug or something isn't working as you expect - you must be brutally clear on what it is you're experiencing, what you're expecting and the steps to reproduce. Your communication is most likely the barrier between the LLM doing what you want.

9. Use console logs often. If something isn't working a console log will help identify it. Luckily LLMs are console log machines and usually always add them when building new features. Be sure to clean them up after everything is working as expected.

10. Don't be afraid to ask other LLMs to help fix your issue if needed. You can get lost in an infinite loop of trying to fix an issue with one single LLM (and want to rip your hair out). I've had some good experience with GPT-5 one-shotting bug fixes that Claude spent hours trying to fix. Try stuff out and try not to go crazy.

11. Go slow to go fast. It's tempting to want to rip through new features fast but the slower and more meticulous you are with the code that is being written the better your final product will be. Pause at every step, have the LLM verify it's work. Then you verify it's work. This will save you hours of debugging and frustration.

12. Don't blindly trust the LLM. I mentioned this before but it WILL introduce complexity and/or bugs if you are not careful. Be meticulous. If you are building something you should want to pay attention to all the details.

13. You will definitely get extremely frustrated at some point that something isn't working the way you want it to and for some reason it can't fix it. In those moments you should take a break or go workout or something. Remember that you wouldn't be nearly this far without AI. Give that mfer some respect.

14. Tell the LLM how hard you want it to think. Terms like "think hard" or "ultra think" tell Claude to think much harder and it will reason deeply on what it needs to do. I'm sure this works similarly for other models.

15. If you're making database migrations make sure to save backups and have a plan in place in case you need to roll-back. This seems basic but if you break something in the database - you will regret it.

16. Security is extremely important and we see cases all the time where breaches happen. Make sure you are building with security best practices (ask the LLM and other LLMs). Make sure you regularly run security audits on your code base. Don't expose API keys. Use secrets. Protect PII. etc.

17. As your project gets more complex you need to be paying more attention when making changes. AI likes to change and/or break stuff randomly. You must watch what it's doing exactly especially as your project grows. Always try to make your changes as minimal as possible.

18. Ask the LLM questions. If you don't know how something works or if you don't know why it's going to approach the task in a certain way - ask it. Ask why it's doing what it's doing and if it's sure that it's the best most efficient way to approach the task. Again, it will add complexity if you don't pay attention. Understand how and why things work as they do and you will be much better for it.

19. Always keep your context window as fresh as possible. This one's obvious but the more clear the context window is the better outputs you will get.

20. Provide as much detail as humanly possible. Make sure it is extremely clear. Do not assume the LLM knows what you want or expect, ever. The more information you put into your prompts the better. Make sure it fully understands exactly how things work in your app or else it will just make stuff up or go off bad data.

21. The more time you spend planning the faster and smoother your implementation is. You are what's wrong when it comes to vibe coding - not the AI (most of the time). You are limited by how clearly and specifically you can communicate your goals and intentions. The AI is only going off the words you tell it. Be ruthless with the details. I've mentioned this multiple times but it is extremely important.

22. Embrace the code. If you are non or semi technical you HAVE to get into the code. Even if you don't fully understand all the syntax you should be able to at least somewhat generally understand what the code is doing. It's important to know how your app is working as much as possible. You wouldn't eat a meal if you didn't know what was in it.

23. Remember you are vibe coding. So just vibe. Let your LLM take the wheel sometimes and get creative. You will be surprised when it actually comes up with good ideas, features, functionality, UI, etc.

I wrote this because I see a lot of people "building" but not many people actually talking about the things that help them. These are just the things I think about and use everyday.

[Alex @itsAlexGuerrero]

I am going to win.

[Alex @itsAlexGuerrero]

Is anyone here experienced with building mobile apps?

What should I be looking out for when getting ready to publish my app to the app stores?

[Alex @itsAlexGuerrero]

@alexfinn This is how all algorithms should be. Prioritizing good content over anything else

[Alex @itsAlexGuerrero]

@alexfinn Good stuff. Have had some pretty good experiences with GPT-5 at least with bug fixed so not sure why it gets all the hate but it is definitely powerful. Gotta try the self reflection.

Have you noticed it work any different when using in Cursor?

[Alex @itsAlexGuerrero]

@karpathy Sounds like an AI product should be able to handle this

[Alex @itsAlexGuerrero]

Make big bets on yourself.

Take risks.

Fail.

Make mistakes as often as you can.

It's the only way you will move forward in life.

There was not a single entrepreneur who took no risk and did anything epic with their life.

In fact the most successful entrepreneurs took way more risk rather than most would be comfortable with.

Risk isn't something that you can avoid.

You MUST take risk.

This doesn't mean you need to set all your money on fire.

But at LEAST take some calculated risks.

Make just enough money to survive.

Spend the rest on making your business grow or investing in yourself.

Take more risks.

Make big bets on yourself.

You will not regret it.

[Alex @itsAlexGuerrero]

@alexfinn I’m on it

[Alex @itsAlexGuerrero]

@AlexHormozi will probably do $100M within 24 hours of his book launch.

He might be the greatest to ever do it.

[Alex @itsAlexGuerrero]

@ZssBecker It is genuinely a marvel

[Alex @itsAlexGuerrero]

@elonmusk How does it make the algorithm better exactly?

[Alex @itsAlexGuerrero]

@xfreeze How is google in the negatives

[Alex @itsAlexGuerrero]

@elonmusk Grok is still extremely underrated

[Alex @itsAlexGuerrero]

��You’re not making as much money as you want because you’re not as good as you think you are.”

- Alex Hormozi

[Alex @itsAlexGuerrero]

@paulg At this point it’s just coding isn’t it?

[Alex @itsAlexGuerrero]

You guys are cooking lately

[ElevenLabs @ElevenLabs]

Today, we’re launching a new Video-to-Music flow in ElevenLabs Studio.

In one click, our Eleven Music model generates a custom soundtrack based on your video’s context.

After adding music, you can layer in voiceovers and SFX directly in Studio.

More examples in thread.

[Alex @itsAlexGuerrero]

@theJayAlto Creation is what humans are made to do. Create value for others. That should be everyone’s goal

[Alex @itsAlexGuerrero]

@xeophon NotebookLM is really good for this use case

[Alex @itsAlexGuerrero]

It’s crazy how these numbers control my life