Vibe coding vs agentic engineering

So it works on both sides the second one is from threat down They release an article just a few days ago of basically putting a 60-day timer or a six-month timer on

extreme amount of AI attacks Right because all the script kitties from before that used to try and hack things now

have been able to make much more nuanced attacks and now the cost of recon and mapping has gone almost to zero and

Especially now with the obliterated models that they're releasing where the model does not fight back and does it as willing to create malicious payloads

It's dramatically dropping the bar for who can attack and what they can attack And so we saw that even I think it was last week or the week prior to that that a lot of large head funds were

attacked using vision which is a form of Fishing using actual AI voice to replicate one of their superiors to then get them to either give a credential or move through

And so we're seeing that across all the big companies, but then quickly that's going to be distributed to use against the smaller middle market

so it's very important that we Work towards finding all the gaps like that's kind of the thing that I put your hair on fire and go find every vulnerability you

Can because in the next six months anybody's gonna be able to go and pick through your stuff and find where you left the door

That I see a lot of nodding heads So I got some people that that are in that in the the space you guys have any opinion on that or feeling

Yeah, and it's like with the shut I don't know how to pronounce with a shahalud worm Right how that went through and knocked out tan stack and then infected into a lot of different places

But all you had to do is actually set up your configuration for NPM or pi pi and you wouldn't run the risk of it

So there's a huge percentage of the industry and especially these new coders that are coming in place And it fits perfectly with us talking about live coding today Is there some very basic simple things that you can do to completely protect yourself from a majority of these attacks?

And I shouldn't say completely protect yourself But you can get to a much better place of safety and that was one of the first Open-source projects that I launched is called VibeSafe Where I built it steps on top of your coding agent and will help scan through packages to make sure that you're not going to

Fall to slot squatting is really what the term is for it where they'll make a package That's close to the same name and they're hoping the LLM is gonna fudge up how they spelled it when they do the import To then pull malicious software into your into your application and into your system and so I built this open source project that sits there and will Go through and just does heuristic checks just like has this been Recently updated is this like been approved is this is there actually an NPM info page there like is it a real package or not?

Just to keep you safe on that side as well as tools for making sure you don't publish Secrets to get help because that's the next biggest and easiest thing for anybody to do is

Publish secrets to get help publicly and then all of a sudden you look and so if anybody wants to play with it

I've got about a thousand people that have used it so far So yeah, oh yeah VibeSafe And then the third piece of news I want to bring up is actually my newest open source project that I just launched called

Which is an agentic software factory I've built it so that it can bolt on top of whatever your favorite coding agent is and

Make it where anybody of any technical level can act applications and it does that by first interviewing with you to understand your technical capabilities and levels and then setting its own standards for how it

Talks to you and how it helps you decide what you should have in the app What shouldn't be in the app does research through all open source projects for good examples of code to bring in and?

then is able to do full end-to-end testing using a few tools and

Brings it to you to just really focus on whether it solves the actual problem Instead of having to think about architecture having to think about anything else And so if you know a problem really well, you can use this to build the solution to that problem

So I just want to point out that normally Justin looks like me was t-shirt and Now the panelist dressed up. I even like to choose there. Thank you

Thank so what I'm gonna do is just kind of do a light introduction of each other and then the follow-up question I'll have is to have you you know The only true person that can really do yourself justice is you so I'll just do a light introduction and then I'll ask guys

to introduce yourself fully so Rashid, and I'm going to

Probably about that week. Yeah, that's right And so it's a company for us. She is actually a member of startups the board members startups act and startups act is a

it's a nonprofit that supports entrepreneurship within the region and startups and

In these been through like he's kind of like I'm gonna let you go on that But one of the reasons I asked for shod the comments he said that his owns developed

A SAS based product Called the space now that you talk more about a bit. A lot of it was done through vibe coding Justin Justin Mendez is Do you want to watch this a principal at GreyDot Media? I'm head of AI head of AI and

You know, he's actually been one of my volunteers since almost day one. We've been doing this for almost two years

2023 2024. Yeah, so And he just had a baby army veteran and

Hardest thing in my life is raising this child He has all of his teeth coming in at one time and it's been a fun three months so far

The the other thing too is that he's you know, one of the things I know just it's been he doesn't have a developer background

But you've been doing a lot be creating a lot of applications to see vibe coding Yeah, I've been coding for about 10 years. I haven't done any courses or any degrees in it It's all just been sweating at night read like watching YouTube videos and teaching myself how to code. Okay, and finally have Orion

predicted it's Weird way we met we were waiting. There was a call for movie extras a few months ago

Then we were waiting in line to get on Lee movie. They shot here Yeah about the gold rush and he wanted to be like the white overlord I wanted to be like an Asian guy that But we said we had like we were there for like a good maybe two hours and we just talked and What I felt very interesting is Orion actually is a developer, but he spends more time now

Prompting and I thought you know, we'd start with you know, getting his

His his experience. So with that, I'm gonna ask you to go deeper So we're sure you want to go talk, you know, introduce yourself more formally and talk about what to do. Yeah, definitely Hi, everybody. My name is Rashid. I I'm a technologist. I've been doing software engineers for the last 14 years Started here at our school long time ago and Grew up from there

This on the side of the space. This is my third company The the previous ones didn't make it past the idea stage and rough

MVP because it was rough So it was very difficult to get a company on the ground and and do marketing everything else

But things have changed for for us and the result as we space we have paying customers

We have a lot of product and we're excited to see what the future brings Very excited to be here and I'd love to learn from each and every one of you guys So, my name is Justin Mendez I am head of AI at GreyDot Marketing where we really specialize in helping

Businesses get found by their best customers. And so my time there I've been here roughly five years at GreyDot Marketing and I've gone from really focusing on web development advertising and SEO to as this AI wave has come really Focusing on that and applying it into real businesses instead of just making fancy little demo repos

Yeah, all right and Berdic I've been doing software for about 30 years and Used to work at Apple and Adobe and worked with companies like Sony and Microsoft from like a partnership standpoint and

First five years I was doing like software testing and then moved into software engineering and design and stuff like that these days

I'm kind of a one-person software company because I built software that got bought by someone who doesn't really do software So I'm kind of the developer designer of several different products that are in heating and air conditioning

But then I have a side project I'm working on and I've always got other interests of things I've got going in the air so my machine has like a million things running on it and I've kind of made that transition from writing every single line of code by hand and like leading teams of developers who are all

Code reviewing everything and being super picky about standards to now moving into having robots help me do a lot of that a lot faster

So it's been a fun process So I'm gonna just throw up the first thing I'm gonna do actually before we go dig it Like can you guys just give me an idea of your coding background like, you know So she yeah, so as far as coding background

I Went to the Community College. That's the Sacramento City College and

America River transferred to UC Davis did the three or four quarters at UC Davis and

Parts full of the job before graduation that thought instead of paying the money for the BS

interview size shift to Sac State and did an MBA in finance. So that's That's a shift from doing just pure technical development to Involve in business as well. I Specialized in Java. I love Java

But I understand also that It's just a tool in the toolbox So as we need tools that we can pick up is great and then with the address of AI I don't have to know the nitty-gritty of the language. You can just tell it what you want based on the

Application and what you want to be doing with it Database wise across the spectrum Postgres SQL and by sequel and all that stuff

And As far as the you know infrastructure AWS is my preferred I do have a

Architect certification for AWS pushing for about

Finish that up But So when I came out from Afghanistan, I went to school at West Texas A&M and originally I was going for medicine from my experience overseas and I figured that wasn't going to be my life And so I transferred over to finance actually in one of my electives was a basic HTML JavaScript and CSS class back in 2015

And so that was really the start of it. And since then it's just been coding at home building projects and it was really 2023 24 where I got to the point where I felt comfortable charging people for the code that I was building and

So fast forward to the day when you actually write and pull yourself. Yeah, 100% by myself Yeah, and so fast forward to today where I've closed and shipped over low six figures in software using Adam

and booking a lot though, so All right. What am I you arrived? So I actually have a degree in audio engineering. So I used to work in recording studios years ago And my first software job was working for Apple testing like co-tools and all their audio applications

And to give an idea of how long ago that was we were evaluating the next OS and some others to see which

Software we buy before Steve Jobs came back to Apple. So that just gives you a sense of how long we're talking about

So I helped test next and so did you actually meet Steve Jobs? I never met Steve Jobs. No, I was an Ireland working

Not that I'm from there, but I was living over there But then it got into work in Adobe and other places working on video and audio software And then when I moved more into writing code I left some of the audio and video behind and was doing whatever product was interesting So I moved more into dotnet and then started doing React and typescript and I've always been focused on the user experience and UI side of things

So a couple jobs ago, I was like, you know mostly focused on front end for a lot of the places

So that's I'm always happy to work on either an API or UI where there's that interface between the person

and I like to I'm kind of a Right and left brain person. So a lot of engineers are either like super technical or Designers are more creative and I kind of sit in the middle So I like to figure out how to take something really complicated and weird and cryptic and make it fun and easy

And you know, that's the really fun part for me So, I mean that just asks us to all you anyone want to take a stab but

Finding so we're out kind of level set. What what is vibe coding vibe coding is the desire to one-shot applications, right?

You're using vibe coding to offload as much of the work is possible onto the robot Instead of taking the ownership of that translation from language and intent to tokens to ensure that it actually builds the product

You know and that's the big difference for an agentic engineer Which is the new buzz term is something that's really invested in solving the problem and also Solving that translation gap between human language and intent and the tokens Well, I think there can be kind of a spectrum because the technical definition if you go from who coined the term It really was I don't know what the code looks like I'm just telling it what to do it's doing it and then it for my world coming from being a coder I don't go that far But I've gotten closer to that because I've trusted as the models have gotten better

There's a point when you're building so much you don't have time to read every line of code and say I don't like The way you did the CSS at a certain point you say hey you wrote a test the test passes I tested it works I generally looked at things that looks okay. So that's the level of ownership. I'm willing to hand over other people

They're like, I don't even know what code looks like. So it can be a little bit of spectrum. Technically I think you're right. They're like pure vibe coding It means I don't even care what it's doing and then there's people that like call it vibe coding But they're kind of collaborating on the code so it can be a little bit of a mix totally fair Yeah, I was more on the intent level more than natural level. So I appreciate that. Yeah I don't know if I can put words in defining vibe coding I have yet to figure that out, but I can bring up an example. I think the example of speaks

Avoidant my wife. She's never included. She doesn't know how to code. She's

she loves arts and creativity And she loves cooking and so what she did that when you know, the tools became available She by put it her way into creating a cooking application. And so to me that's like so like you you don't

necessarily need to know what the code is written how it's written what it does and what it doesn't as long as like you said There's the final outcome is exactly what you have in life So, you know when I've talked to developers about vibe coding, there's a few that like Like don't don't only want to touch it, you know And I'm kind of curious what your first experience and you guys could answer what your first

You know your first thought when you heard by coding and like what was your first experience in vibe coding was?

Yeah, so my first experience with vibe coding was back with GBT 3.5 turbo Right because my main limitation is the accuracy of my fingers and the speed I can type and not matching up with how fast

I'm thinking and that causing a major issue. So as soon as I realized that even Putting my info into chat GPT 3.5 turbo and really being able to pull something out

Put it into the application and run it on the dev server and it worked. I was like, oh my goodness now

If you look and think about that that sounds almost like caveman style compared to what we have today But at that time it was like, oh my goodness. I can build to the size that I want to I'm not limited by my

own physical abilities At that era though because I started using things then and it was cool, but I used it to solve like a specific problem

Oh, I've got this cryptic error No, it is yelling at me because I try to install a package and there's a conflict and I don't spend an hour figuring out What the heck is going on? So I'm gonna have GPT look at it. I'm gonna paste stuff in it's gonna give me something back

Try this command. Try that command. So it was fun But it was kind of like having someone you could call when you have a problem. It wasn't the same as hey robot Here's my code world Go build me a new page that does this or go fix this bug or just iterating on this page until this error goes away

That's where we are now so to me it's the last six months eight months that The the tools have gotten good enough the models of that is good enough that I can give that all my code and say

All right, you kind of own this now add this change that that's where it really to me spilled over it to okay

Now I can do five times as much whereas in the GPT 5 era the 3.5 era It was it was a little more like here's a message I'm giving you and there was no it didn't have any control over your code. It couldn't really see all the contacts

So it was very limited at that time great point to make and really like one of the main value drivers of AI is

Accessibility right like everybody uses it to search the internet and to go through a lot of sources I would use it to look through stack overflow for me because I especially at that time I wasn't really Knowledgeable enough to know if that was the right solution to the problem And so it would be able to look through many of the entries in stack overflow bring them back

You know, this is a that's a great example one thing that I think And it goes back to to your ability to do both the UI build-up and the backend

I don't I'm not very good at UI design. So One thing that I use my coding for is do research Go to research come up with the best use cases for a SAS that is being used by this

By this demographic, how can I impact them more? How can they like like it?

So the UX needs to be ideal it goes and it does like you said research

Paper what paper research not just random guesses You know with the with References and so on It comes back with the best design decisions that that I can actually implement and then I use that to

So I think it's very valuable. It says a lot of time During my thesis. I spent a whole year doing research And you know ended up writing a hundred ninety nine page document

We spent 20 20 minutes with chat GPT and it goes through libraries and libraries and libraries and comes back with

But you touched on something really key which is your you review the research it gives you right one thing

I think people have different experiences with this How much do you trust like the AI is going to give you information with that information you have to decide is this accurate?

Is this all the data that I need and same when it designs something for you the taste and the judgment?

Is where you still have a role, right? Like you're not at a point where AI can just make me a company and do everything design it build it test it

Support it there still needs to be somebody judging. Well, what idea is the right idea?

Did the solution that it proposed make sense? Is that a secure way of doing it and you trust it to some degree? But you also take your experience and your knowledge and you can also ask it like hey Where did you get this information from? Can you cross check? Can you hey this AI? Can you check what this other a I said and tell me if you disagree and that's where you have to kind of use

Whatever is at your disposal? So I think that kind of touches on what you're talking about where you're not just having it write the paper and then it ships It publishes it online right here You're looking at that and deciding like do I get what this means? Does this make sense to me based on my experience?

Yeah, it's a very strong point because the input they give the AI

And I'm pretty sure we all know it. They're all I mean all of the one nice to you. You're absolutely right

And it shifts and and kind of molds to In response to our input and so it's very dangerous to trust it blindly I remember a case where one customer they turned Immediately, they they entered the you know, the credit card they locked in and when they created a property

They clicked on a flow that I haven't tested. So they they hit a wall. They couldn't move forward and I lost them forever. So

When I went back it was that kind of interaction that I remember where I I pulled it away and it became sort of

Two people who are not in alignment in agreement and the result is catastrophe. So that's exactly what happened

So yeah, don't trust it blindly. We're not at that stage yet, too

Yeah, and so please correct me if I oversimplify this at all But like if you look at the nature of what an LLM is, it's purely statistical inference, right? These billion-dollar companies have put a lot of marketing in to tell you that it's thinking that's not functionally that what's happening, right?

It's predicting what the next token is in a sequence It's not actually putting those thoughts together now when you see in those the UI it says thinking really what's happening behind the scenes

Is that it's running additional tokens and trying to optimize your prompt to get the most likely answer there

And so if you're dealing with statistics, you're always going to have edge cases You're always going to you cannot have that stable same deterministic of basic code

Right. So the statistical part of this is a huge area of debate And the interesting thing about it is that it's just a sophisticated autocomplete. You'll hear people say that on tiktok, right?

Oh, you're trusting this token prediction word prediction thing and there's a there's truth to that and how it's trained

But when it's answering a question if someone tells you they know how it came up with that answer, they're lying

Yeah, the truth is the people that build AI don't know how it answers the question that it answers They've given it all this if they've given it the corpus of all of the information We know and they've given it a way to train itself after that. They don't know how it works

Exactly, you can trace kind of neuron pathways basically that are being created in the software to see how it figures this out

But it's when it's answering you if you look at what it's what thinking it's doing you read the output you realize it's not just randomly predicting words

It's actually Understands the concept of what you asked for and it's trying different ways of doing it and it's coming up with an actual solution

Instead of just saying oh he asked me what animal I like So I'm gonna say chicken because that's what most people would say, you know That's what it sounds like when you say it's like predicting an answer But the truth is, you know They've done tests with AI where they can they can give it a word in different languages and say what does this mean?

And it understands the concept of like what Brown is regardless of a word Right it actually uses it knows that Brown is a color and it knows what a color is without the language, right?

So it's not necessarily just predicting work I agree with you there But to think the key here is that it's not actually processing words, right? It's processing factors and that's about cementing the semantic like reasoning or semantic

Like of what Brown means in multiple language semantically. It's the same thing regardless of the actual characters

That the encoder and the decoder put it into, right? And so that's that's where I feel like truthfully it's taking vectors and doing matrix multiplication through the system

We don't know because it's a M forest or a blind forest now See, this is where I get limited on that deeper technical understanding, right? But the truth is it's math that's happening, right? So it is pure as we should all look at ourselves and how we think that we're special snowflakes

And the truth is we can really replicate ourselves using an extreme amount of statistics

So I'm gonna go back. So we started Orion and you're like a one-person development shop for your company

It's an HVAC company as I remember, right? So I think you just said 5x. So now that you're using vibe coding

Are you essentially your output is 5x before you 5 code? I would say roughly. Yeah

I mean some sometimes it feels like 10 sometimes it feels like 5, you know It depends how you quantify that but when I actually look at the weekly update I send to my boss and say this is what I built It's five times what it used to be and not only that but I'm building a bunch of tools that help me do things

You get away? Right, but like my feeling is it's my obligation to my boss to say well These are the tools I have I'm gonna work just well Sometimes I work harder now because you feel obligated to like I gotta use all the robots that I have available, right? So I'm sometimes I'll do more but yeah, it feels like I'm outputting that much more stuff And I'm also writing a thousand times more tests. So I'm putting in more checks I'm auto fixing stuff that pops up in century because I hope some user out there got an error somewhere

So now I'll just automatically fix it instead of putting it on the back shelf somewhere. I was like, I might as well fix that I've got

Agents available to do it. So you do you find yourself doing well, you know before you need old top olden times

You were coding now you're more prompting and like doing more QA Yeah, more design. Well, it feels more like you've got ten engineers doing whatever you want

Like it feels like I have in my garage. I've got ten robots standing around going I'm bored What you know, what can I do you paid for me? I'm standing around please give me something to do You know feel almost like an obligation to feed this tasks So you come up with stuff that you're like, well, if I just had unlimited resources What would I do with that and that's kind of how it feels now as I realize there's so many things over the years

developer that you say well sure I'd love to have 100% test coverage and I'd love to have an automatic bug fixing thing and I'd love to have

I'd love to translate my app into Spanish and I've just never taken the time to do it And that would be pretty cool. And now you just go, you know what? Why not? I'll just say hey make a Spanish version and it does it and you test it and you go, okay that worked So it opens up a lot of stuff that I and you start thinking differently

So there's a lot of things I do day to day that I've made little tools to make it learned how to clean up a lot

Of old junky stuff that was in my workflow you talk about AWS I used to have servers on AWS that I could terminate but I couldn't delete them because they were tied to some

Rule somewhere in AWS and now I figured out how to delete all those You know, it's sometimes it's little house cleaning things But it's stuff that just makes me more productive and makes me feel better every day as I'm and then

What kind of like have you been modern in your AI cost?

Yeah, I mean I only spend up till recently I was on the $100 Claude max plan and that's all the AI that I buy I don't buy anything else

I had pay for aqua voice, which is like a tech speech to text things. I talked to the AI on the day

I'm Apple's the one that runs on my phone and was cheaper than whisper voice or whatever. So I pay for that That's like $8 a month And I recently upgraded the $200 plan because of fable. So now the fables available But you only get 50% of your usage on fable and I found it was way better at design It was way better at architecture. So I thought all right, I'll upgrade for a while But I probably won't stay on that level Well that reason for the issue question what what AI tools you guys use I use Claude I use

bedrock when with the AWS I use GCP Google Gemini For startups guys, if you have stuff going on AWS and Google and all those guys they have the startup programs and you can get free credits

In the thousands of dollars. So very interesting if you have a startup in place of by

I use cursor Which allows me to use all of the different Frontier Labs models and often I use them together at the same time

My budget cap is at $1,000 a month But now that XAI acquired cursor and pushed rock into there and do poser models are doing

Fantastic work. I've pushed 2.6 billion tokens this month and I haven't gone through my two months subscription

And then it also gives me $400 worth of credits for the other Frontier Labs So I use favor a lot Or the other models codex Gemini because they each one of them have their unique kind of personality and strengths to them based on

What the data corpus was available to that company? Then you have the tools that you're using. Ah Claude code is 99% of what I'm using I do have a deep seek account for API usage because I have a side project which is a language app I'm building like with learning app and Claude's API Tokens are a lot more expensive than deep seek. So I tried deep seek to compare it to you That's pretty much how I do I mean I'm spending like 10 or 12 hours a day and all I use is Claude pretty much for all of that And I've moved half my work on to on device models specifically on Apple silicon using MLX

On especially a quen 3.5 of the 27 billion token model that was just released is extremely capable

And right now my laptops in the car But at home my MacBook Pro is running the live code bench

And I was going to do SWE Lite to really test my agentic harness or coding harness that I built myself to see if I can

Get up to those same benchmark levels You know, so we're kind of like living the dream right now. I have so You know, it's like the the one-person startup So tell me like how did this how did you start? Yeah, so the the It started because I invested in real estate bought a couple multifamily properties

And I had a property manager To manage it. I was working full-time. So it didn't make sense that I'll be Chasing make its requests and text messages at night I had the property manager realize that they are charging they're saying it's 10% But it actually when you do math, it's 16-17% It's a lot of money right? Yeah for for the the income or the revenue So I thought okay, what would it take to self-manage?

And having done software, I thought I'll do my own tool and AI helps quite a bit. So to get me to market pretty quickly So you you basically use vibe you're using that money to solve a specific problem for you

Yeah, but then when did it when did it when you decided to like, oh I could actually market this and

Yeah, SaaS product. Yeah, so I as soon as I created it I realized that okay, there's some costs. I'm gonna put in somebody some time

I need to monetize it. Yes 10% savings of 15% savings is great But at the other day, it was destined from day one from the zero to be

You know a SaaS that I'll use myself on them my first customer. So that was the idea

problem solved in my own and You know, I I feel like this is a brand that's in my kid

You know Vibe coding has this Potentially negative politics. Yeah, and And it reminds me of

Earlier is reminds me of early days Wikipedia

Not that great but then it evolved against all social resistance and all into a reliable source of

I think that's where vibe coding is going. I think it's You know, it's inevitable Just because of the sheer number of hours and money and everything that can save

Going back to what about this up is a long time ago We couldn't imagine starting a company and be a one-person developer marketer Salesperson all that stuff

But now it's cost on the living experience of that And so with the time savings with with cost savings, we see it everywhere, right? There are

Not necessarily because they're not going to have a job later they're just going to be doing something different but

I don't think you're going to be doing the same It's not going to be business the same old way. It's going to change. It's already changing

so Adopted embrace like cool. It's then it's the next Wikipedia guys

And to your point here and this is my side com for I actually think you know

There will be jobs that will be lost today I but I think what this what's going to happen is it's going to create a lot of entrepreneurs, right?

So, you know, anyway, that's kind of what I think is You know the future is like gig work and entrepreneurship So if you ever heard of the coastline problem

So if you ask a scientist or somebody that's very precise with measurement and also other coders as well

If you try to measure a coastline You can only get an approximation of it because the more you zoom in to get the accurate measurement of the coastline the more nooks

And crannies that you end up finding right and that's what's going to happen here as AI adoption happens and the abilities

Raise for everyone based on the access to these tools It's going to make these unique cuts in these unique places where everyone's going to be able to hyperneish into solving one specific problem

But being able to then serve that to the entire world because of the support that they have

That's my prediction. Yeah, and my my Person owns the company who I work for that he told me he feels like over the next year as the tools have gotten So good, we're going to see a lot of people Bespoke making their own stuff, right? You're gonna people are gonna say well I can just build my own little inventory management tool instead of buying this one or I'm gonna build a tool that does this or that

Just yesterday I built like a memory management thing for my toolbar in Mac OS to tell me like how much RAM am I using and all that because I was like why buy one I can tell it exactly what I needed to code it in like five minutes But there will get a point where a lot of those people realize I don't really want to do this That's my job. I tried to do it. I tried to sell it I got 40 people to use it. It was a bunch of time and effort. I don't really want to do this anymore I'm not a software person I'd rather just buy one for 20 bucks a month or $100 a year or whatever

And so I think there may be a surge of like everyone thinks they can build software and then maybe that'll ebb

And then maybe as the tools get so good Maybe at some point you can just prompt it It'll make something in two seconds and maybe no one will have to hand do any software anymore But it is a big debate right where we're all going to land and all this and this has happened to a lot of industries

me coming from a music background in this 80s 90s if you want to record music it costs you half a million dollars to get a recording contract to pay for the studio time and the producer and the engineers and musicians and Then now we have like Logic Pro on our laptop So you can buy a mic for a hundred bucks and plug it in and make an album And you see people doing that people having number one songs that they recorded in a hotel room with a laptop

Did that mean that music isn't an industry anymore? They don't have to record the music anymore

That's a whole other question, but but you can see how that Democratized like the access to the tool, but they didn't make everybody a songwriter They didn't make everybody a singer, you know It made it easier for that independent artist to make the thing But then you have to figure out like how do I get people to listen to this and that's user acquisition I think it's gonna be the biggest challenge you could there's a lot of great products out there That nobody uses because you they can't figure out how to convince people to give it a try

And so I think that's gonna be the real challenge where everyone's gonna make stuff and go wait I made this cool thing why how come only 30 people? Paid for it and you're gonna realize that's the real hard part That's why Instagram got sold for two and a half billion dollars. It wasn't you could build Instagram in the 45 minutes

Yeah, it was the fact that they had 36 million users. That's what Facebook paid for

Oh No, great great call back I love that you bring this up around and that's like when everybody realizes that they can use the AI to build things and they

Get super excited about it And all of a sudden you're gonna try to go in 75 different directions and try to build all your own things You realize that the uptake and the management the updates everything there has a cost to it And would you think about when you're looking at these subscriptions?

You're like, oh, I'd rather not pay the $50 a month The truth is with that $50 a month You have a whole team and a whole company of people working to make that product better We'll keep it up to date manage it everything for you for just that $50, right?

And so that most scarce resource now that we have this ability is our attention So it's finding that one place to really focus on and keep going and that's I struggle with this the most right?

I've been diagnosed with ADHD, but I love to have 75 things going at the same time

I'd love to have that level of stimulation and I've got myself in trouble with it And now I've locked my my cursor like in the rules and the hooks where I set a focus for an entire week

And if I put an entire week on a single project the amount that I can get done is amazing Right versus having 75 different chats in different projects running all at the same time. That's fun It gets my head going like it does all the same dopamine jacking like being on Instagram or anything else but the truth is to stay efficient and to get to a product that you can actually sell you need extreme focus and

That's actually the problem that I'm working on solving right now with my own startup project called Arthor Which takes from I said GreyDot Marketing and helping businesses get found by the customers every day I've distilled all that information and all of my SOPs everything for ads SEO web design

sales Into the rules and skills of this multi-agent system. That's gonna help

Yeah, AI slop ads to that right because like my boss just told me two days ago Can you make more videos walking through the software because now we try to make it with AI and then no one wants to watch them

Like we they cloned my voice and they were trying to auto make videos with my voice and then people would just be like I don't want to watch this So he's like can you start recording more and because that's what people click on and so there's always that now

We've got so much content that people could spit out and people are recognizing. What's What's AI and what's not but it's not just what is AI it's What is something that people put the care and the taste and the judgment into to make it good? Because you can make something good with AI. There was recently this 13 minute. I'm not saying this was good I'm saying it's an example of what you not to do. It was this horror movie that somebody made recently some filmmaker guys So it's like 13 minutes long pretty cheesy, but technically it's impressive, but it's not a great movie

But if Christopher Nolan sat down with AI and made a 13 minute movie I bet it would be pretty good right if he sat and really cared about it And we believed in that he could probably make a good product So it's that's where I think you come if you combine that you can still make something good. It's not the AI that's bad on its own

It's that if you have someone at the wheel who can't judge what's bad and what's good And that's why you see all these ads now. We're like, how did they post this? This is terrible They just prompted it said okay, and they posted like a thousand

Yeah, so that's where the slot really creates a lot of noise. Yeah, you got to make the sandwich, right? It's human on the first five percent. It's human on the last five percent The ninety percent messy middle you can give to the AI as long as you really invest on putting good quality tokens in and

Defining what correct actually looks like right through testing because that's everything I do is it has to set red-green tests

And it watches another sub-agent work on everything bring it back to get tested and critiques that work

And then finally comes to me at the end when everything passes green to then me go through physically through everything myself

Well, I hope is this this raises ideas To a higher level than just who has the money to pay a big team to build something because a friend of mine I just talked to about he's a web developer. He doesn't really code I was teaching him how to like to get the AI stuff going and he said well What if I build this product and then somebody copies it? How do I like copyright it?

How do I trademark it and I'm like you're missing the point that your idea. I'm sorry to say It's not what's that valuable Someone else can literally look at a website or an app and say I want to make something exactly like that And there's no legal recourse to say you can't make that that looks like Instagram The Instagram can't stop you from making a photo thing that has all the same stuff. There's us, right?

It's the point is the users and also how well it works How what who you partnered with to get it in front of people so it again democratize is one part of that

Which is people before we said I have this amazing idea, but I don't want to get investors I don't have a million dollars to pay people to build me a prototype now Now we're gonna have a lot more of those people fighting for the attention Yeah, that's a great point when a lot of people When a lot of people create apps the difference is who takes it past a certain critical

Point first level second level third level a lot of us will get to first level and then we'll stop

How many people use lovable and then they got to a point where they got frustrated with it

And they tell it to do something it's completely something completely different Yeah, I know a lot of people who do that they get to the point where The system built something in the background that is not scaled and so it hits a virtual ceiling

And it's not gonna go past it so Yes, it's good to let it rip and take care of itself

But the decision-making I think should stay with it. We should architect it We should follow the the normal process Who am I building this for? Is it for 10 people? Is it for me? Is it for a hundred people?

Is it for a million people? That will determine whether I should architect it spend some time on architecting it because the blueprint

AI is not gonna do it for me. It could you could use it to help with the blueprint So even that step the evolution step of the building system, you can use AI to help you with it

But you are the driver like I wrote AI wrote many percent of my code base, but I made a hundred percent of all the decisions. I'm architected it

I know where I wanted to deploy it. I know how I do it. I know I have 18 micro services. Each one does a domain specific

Area, I decided that and so when you architect it It does have the potential to skip versus you build something and it hits the ceiling and you don't know why AI is

so stupid That's a great point that you make machine because you that you are right there next to the problem You're the closest person to the problem that you solve Right, right And I think that's the core thing here is instead of trying to go find some random market to make some random software for

Find a real problem in your life that you actually can define what good looks like That's the core of making that product right and like you said of lovable these other ones. They're so fast

These alums are so fast nectar because they're they're optimized for fluency more than they are optimized for accuracy

Right and they're gonna Over engineer way before you even have an MVP way before you even get it front customers and get that real feedback to actually make

The product where it should be before you go to the level of production Right. Well, and one last point of that whole thing is that I I'll make a bold statement Which is I'm one of these people that almost hopes that AI gets a little better and then stops And then the reason I mean that is that I'm in that sweet spot We're like, I love what AI lets me do now, but I don't want it to do what I'm doing You know, I don't want it to get so good that my boss says well Why don't I just prompt when I just make you as an AI and have it just build companies for me and do everything

Like if it could judge what I could judge that had the experience but have the taste and the kind of ability to look at

That stuff then it would start to worry like what are we all going to be doing? Right if AI gets to that level

But I still think all these points were bringing up about right somebody still needs to look at this and say, okay Why are we doing it? When do we decide whether it's good? What is the whole purpose behind this because the AI is not very good at that stuff at this point And it's not very good at piecing all that together and saying okay now what do you actually do with this?

You can't really make just a whole company I mean I think in a couple years there probably will be a website where you can say Make me a guitar tuner app go and it'll just build it publish it set up payments set up

You know support maybe it'll literally do everything but you know, that's that's the hypothetically possible

But it does kind of take all the fun out of the whole process, you know, so yeah, there's there's the

You know the society and how people are seeing it to your point we get defensive on the job

Is this AI coming up to my job and you can see it within teams Just six seven months ago. I was a director of software engineering at one of the companies here There is a social resistance enough there is you know the what we spoke about the Negative co-notation to vibe code in and people use an AI but then hide in it so that they don't come across as oh

Yeah, I did this. So why do they need me? When when I try to push initiatives and she seems transformation a huge lock

Who's the resistance and it's it's everywhere Do an interview process when we when you speak with the hiring managers

Generally mid-level hiring managers. You do see the change in turn when you go heavy on AI use

People don't it is a huge resistance. I don't know if you guys find enough, but there's definitely a resistance that being said

The decision makers the drivers are the CEOs and executives and so on So the bottom line is what's most important That's why I'm saying I don't think there's there's a way around it. It's gonna happen. We don't we like it I'm not we just need to embrace it more Incorporating your workflow and For you that's fine, but I gave it the decision I'd be executive and we drove it through

The path to get to the final point and that's the value. It's not the

The the you know repetitive typing knowing syntax and so on it's

The target and the ability to take it to be finished That's a great point and yet so like especially on that pushback side because we see these large

Corporations do major headcount reduction, right? And then all sudden there's an article a month or two later, right?

I think it was famous with meta where they thought that they were really going to replace the majority of their staff and then he's like Zuck said oh, I regret that decision Right and you see you see these large organizations are optimizing for the quarterly reporting Versus trying to optimize for the maximum value that they can get out of their staff It's about empowering your employees with AI not replacing them because if you can take away the repeated like the very

Redundant work that they just do this the exact same way every time and let an agent do that and they sit there as the

Curator to be able to say this was good or was bad or this with these change you can Amplify their abilities like this is what I do it GreyDot Marketing for ourselves for our clients They used to take me three to five hours to do a full audit across all their social profiles They're SEO their website and now I have an agent do a majority of that work for me and it takes about 15 minutes

I still sit there put in the input first of hey, this is a potential client talk to them about these specific things They're focused on goes and does the research Comes and brings it back to me And then I still have to go through and fine-tune the report before I actually send it to them to that potential client saying Hey, I think this is how I can help you, right?

But it took me from three to five hours of work to 15 minutes So it's made it where we can massively go and speak to a lot more people and a lot more revenue

You maybe they could have gotten the metaverse right if they just kept all this Had them all do better work Still a bad idea, but maybe they could It's really simple math, right you think about a company that's got a thousand people you think about a company that's got a hundred people If all of a sudden there's this company with a thousand people cuts their staff in half because they're going to use AI But they still have their old systems in place They're not going to get that same leverage versus that hundred-person company that now is fully adopting AI and they got a 10x increase

So now that hundred people is putting the output of a thousand versus this company that's got 500 and maybe they're putting output of 750

Right. And so it's it's really about empowering your employees instead of getting rid of them. That's why we're looking

That's why we're lucky as people working in smaller worlds because I think we're in that perfect sweet spot, right?

If you're at a big company my last job, you know, it was really difficult to do anything. We worked with banks We worked in complicated partnerships and back then if I tried to bring a bunch of AI if it existed at the time

It would have been difficult right because you're dealing with a lot of interconnected teams You've got a lot of legacy software that has to work a certain way So a lot of people can't, you know, a friend of my works at Audible, he's a director there He said they don't use AI for any production code They only use it for test harnesses and tooling and stuff like that But they're Amazon's very afraid of like pushing out a bunch of production code So they went down for like a day because of AI Right. So they had a big outage and they're gun shy about it

So they're slowly kind of moving toward that point But that's why like they can't suddenly make Audible ten times better because of AI and that's the weird thing

I'm wondering about is I have you guys wondered like where's all the great software that's supposed to be coming out Now that we have AI cuz like my son uses steam, you know for video games and I try to use this software

It's horrendous. I'm like, why couldn't you just rebuild this entire thing in AI and make it a lot better

But companies are the incumbent people that have this software. They don't know how to do that They've got too much intertwined They don't want to try to start over with a clean slate and rebuild things And they don't even know how to use AI to make what they have better That's what I'm wondering is like, where's all the improvement? We're supposed to yeah, I see active degradation, right?

Right, you see it off being a way worse. Yeah, yeah, it's like oh you've got to the cookie here and it's showing me a Mexican restaurant

You guys's algorithm is not functioning. Correct, right? You'd think by now We would see like more cool stuff everything getting way better. That's the thing I'm not seeing so that's the thing that makes me less scared that we're all gonna like not have jobs Because I I don't see the C's rising in terms of quality and quantity of the basic software tools out there

Yeah, that's a great point because AI is really good at the what and the how It's not yet able to distill the why behind the decisions that we make, right? And that's that unique experience you have as an employee Like you said with the banking area and like a lot of that was relationship and a lot there's there's so much more inside our heads into our decision making there that we've been not able to extract and Actually present to the AI to possibly get that better generation Yeah, and I guess if anyone's ever worked in like a company that's kind of old-school whether it's like

Government work or like hot, you know healthcare banking I've worked in some of these areas sometimes and sometimes you feel like why does everything move so slowly like why is it so hard?

I feel like I'm walking in like sand and you realize it's because of all the processes and all the incumbent

You know the technical debt and all this kind of stuff So somebody who's really brilliant and productive who walks into that world Can't be a super productive person because they're just too constrained So I think with AI it doesn't solve it for those people You still have the same restrictions that are outside the world where AI can make a difference. Yeah, that's that's a very good point

you know, there's a There's the impulse and then there's the time to reaction and then there's a final reaction. I think we're in the

the propagation point where Where the change happened AI is here We are adopting across the board, but then the bus has 500 people in it the example of the car

So it's about seeing those 500 to all synchronously turn at the same time and not losing anybody across

You know like as you move it's very difficult to

Implement change change management is ridiculously hard to implement And I think that's where we are change management right now. We're companies They're trying to get it AI they are doing it They're pushing hard for it proof is the numbers of my off kind of thing and the technical

adoption of tools out there But it's difficult so it's gonna take time. So For all of us here. This is a perfect opportunity to get a start up going I can squeeze in we can jump in and we will be extremely competitive because they are slow slow to move

We're not we're one two people companies people so we can be versatile

So I'm gonna have to ask one last question before I open up to the audience What advice would you give to a person that is going to strike by the pudding?

Understand the business. What is the business goal of your of your of what you do coding is great

typing is great seeing something become a tool after you just told it to become is fantastic, but then

There has to be a meaning there has to be a higher calling for that effort

And I think that's that's what's going to make the difference between a product or a hub That's great Exactly that is find a problem in your life that you would really have a better life if you solve it

because now you have Tools to be able to build that solution to solve it And if you have that problem a lot of other people have that problem It's very easy for you to speak to that problem across social media and everywhere else to get more people to download it And use it. That's the core thing is find a problem in your life build a life that you actually want to live in I

agree with both those and one thing I'd say is don't believe either extreme of Opinions about AI because you're gonna see people that are like hey, I can't do anything I prompted it to do to write me an email and it wrote garbage AI is useless anyone who uses it as an idiot you'll see those people like flooding the internet right and some people make a lot

Of money pushing that perspective you see them making tons of videos about it And they get a lot of traction and the comments are filled with people like yeah, it's true Don't believe those people, but also don't believe Sam Altman either, right? Yeah, don't just listen to everything he says and go. Oh, yeah, everything about this is like changing the world He just recently yeah He was saying used to take three months to build an MVP for a startup now at 17 minutes And I'm like, I think that's that's a little bit of exaggeration, right? Well, yeah

So you you get people that are so sycophantic that they think like this is this is gonna solve everything and I trusted it

That's a bad idea. You know, believe the people that warn you that like it's gonna lie to you. It's gonna get things wrong

It's gonna glaze you and tell you your ideas are brilliant Those things are all true But that doesn't mean it's trash and then it's useless and that you should ignore it because there's there's a lot of

Motivation for people to go to one of those two extremes Listen to both sides and then find your truth somewhere in the middle because that's like don't don't drink the Kool-Aid

But also don't spit it out, you know, just like give it a chance to rely so much on it

That it's gonna get you in trouble. All right. Does anyone have any questions? I hate to say it

But by the time a book is printed It's like it's so far out of date the concepts, you know, yes, you could get across the book

I personally didn't read any but I would love to have like a tome that some source there there are books

I'm sure that talk about the concepts of what a is really doing but in terms of something that's like

Relevant to today and what things are capable of a lot of times you hear opinions about AI and you realize this person's using a model

This must video must be a year old or this must be you can even tell if it's six months old They'll talk about flaws of what AI does and you go. Yeah, it did that to me in 2015 But it doesn't do that anymore and they're still convinced. No, no, no, that's what AI does It doesn't know how to handle this. It's like yes. Yes. It knows how many hours are in strawberry I don't know how many times you're gonna try to use this But it's it's gotten past that I was 2013 guy or not 2023, right like Yeah, they've got all those things. They're figuring out all that stuff

So so books unfortunately like if they cover the concepts But in terms of like I think you might find some good creators on YouTube That would be a good source probably as people that you you like the way they think you see what they're doing

You know, those people might be good sources of information A book that I would recommend is DDIA right designing data intensive applications, but that is much more at a higher level

I actually just published a full course on my website for free. You can just go to justinmendez.ai

I collapsed my 10 years of learning of how I got to where I am to a 90-day sprint That uses all three courses that I've collected across the Internet and that I've learned for myself Because like the team at land graph knows graph agents much better than I do right the hugging face course

They know all that much deeper than I do So I just I collected that all together and laid it out and explained why you'd want to learn from that

And I also collected the youtubers that I originally like first learned to code first understand the difference between client and server

All of that and I put them there on the page

Also, Nate B Jones is a guy. Yeah, Nate B Jones

He has a sub stack you have to pay for I think but he has Constant tiktok videos and he's someone who has a good job of like he talks about what's good about it He talks about what's bad about about it. He's always up on the latest models. He has a lot of really current useful information he's very entertaining to listen to So that would be a great source of Well, he just mentioned one I think I would be wary of a lot of ones you see advertised

Just out there because I think there's a lot of predatory people, you know You know you find a lot of anything learning how to how to do day trading learning how to sell ads on Facebook anything

That's like um, you know a potential money-making area. You'll have people that just slap something terrible together

So if someone you respect you've watched online or something if they recommend something then I might be a good idea

I think the other thing is trial and error Just try it and then actually now if I don't know how to do something I'll just ask the AI. Yeah, that's

That's what I tell my friends who I try to get into it is like if you're asking me something then you've missed my point Like you should be asking AI how do I set this up and and don't always tell it Hey use my GitHub account to do this say hey find me a source control That would make the most sense for the project I'm doing and have it give you an idea and don't just dictate everything because sometimes it might suggest some good options

And Google's been teaching for a long time, right to use their products They've launched some great courses that are completely free And I really encourage you Joe to go to my website justinmendez.ai. There's that free course I've curated them all so there's links there for you. You can go to all of these courses. I've collected three courses

So it's no charging. You don't have to give an email. I don't want your email. I don't want your money I'm just putting this information out so anybody can get to the same level

Put the list in the URLs for you and how much they're worth because there's a lot of them

For all the cloud providers, okay Yeah, I was gonna ask you the same thing yeah So if you once you start a start you have a startup you probably need a bank account you probably need

you know all those basic services of services choose Two services who have perks with AWS with Google and with with the other ones so

Banking wise I'm banking with Brex Brex has and it's free, of course So you just open an account with them you have access to perks. I have I think I got five grand

From AWS. I got $2,500 from Google. So I use Gemini using those credits. I think I have a whole year free from the data dog

for you know, logs we reported and I have

So a lot of the programs also require like a startup accelerator

So we run the startup Brian and run the starter prime chapter in Sacramento. So

So you could put it you could see what programs are associated with startup Brian Sometimes they may need the director which would be mean to verify that you are part of the start of grind like

like writer What a call out, right? Yeah, I know. So starting up some grind. Yeah Yeah

So, uh, what was the other day on the states? It just uh, oh In October, we're gonna have a Department of partner managed from AWS come here. So they'll probably talk about

Yeah, so I believe we were talking about the new watermarking or Claude is doing

Right and they're doing that just by using specific words and phrases throughout the text And so when you find that repetitively through the text, essentially they're saying that you can guarantee that it was done by it, right?

Yeah So not not necessarily plagiarism The bigger the bigger concern here is the inability to copyright your work because it was generated by AI

That's the bigger fear right because there's there's the academia side of people getting caught for using it to help improve their articles

But the bigger side is that now when you go for copyright like anybody could argue Oh, you just had Claude make it for you. So it's not really something that's your own work So you can't copyright it or trademark it. Well, it's a lot of AI books being written, right? Yeah, nowadays people spit out like write me 10 novels that I can sell

So

So the court cases are still So there's the concern there's a few things like We talked about earlier was a guy made a car AI generated comic book So the graphics and he submitted for a trademark Initially got it was granted and then when they found out it was generated by AI they discontinued I mean they they narked it. I mean they nuked it but The images was not copyrightable because it was generated by AI but the idea of the comic book character

the writing of the story was was copyrightable and I just gave a workshop last last month on

creating AI movies and Jacob Was asking but he says when he creates a character what he does first is he hand draws it

And you know, he like hand draw the character because he wants to be able to if there's any issue about copyright or

IP he has that proof that he hand drew it and then you know He refined it later on with AI the original idea was handwritten So there's there's in fact, I am I actually will have an it's been a while since we had an IP

Lawyer come in but I'm gonna have an IP lawyer talk about AI

And you mentioned IP I mean When first AI just broke ground I guess and became

popular The argument there is that if you use AI you're you're cheating But if we look at our

Life and how we start we're not born with the knowledge that we have we learn it from people around us

We learn it from teachers throughout schools. We learn it from mentors and so on So whatever idea we come up with now and we think that we owe it it fully through and through is that copyrightable?

I I argue it's the same case. It's just not a much faster process Well, and mostly I think people misunderstand copyright and trademark law Anyway, people think that you can trademark your website and no one can make a website that looks like yours And that's not really how copyright works Like someone can someone can't impersonate you and say I'm Pepsi, you know Like those are things that are obviously clear trademarks, but with copyrights a lot more difficult, right?

Like if you copyright a song you wrote and then you say hey if someone records the song They owe me a licensing fee that's clearer to prove but with a cookbook. You can't copyright a recipe for example

There's no such thing as that's my recipe. Nobody can put that recipe in their book You could copyright the cover and the art and maybe the style of writing or whatever But someone could still imitate it and it's very difficult to sue that person and get money You know, they could just say I made one that looks a lot like yours, but it's just different enough, you know with music

I think it's a lot trickier because I'm a songwriter and musician and I record various instruments and there's that debate for me

I would never have AI write a song for me write lyrics write melodies But if I play the drums and I can't play drums very well And I say can you replace these drums with the same idea but do it perfectly and AI does that for me?

I still feel like it's my song But it's sort of like is that different? It's not much different than a drum machine But it's a little different because I'm relying on other people's drumming that was done over years and years, you know

So it's ethically a little bit different But you know, there's that line with music that's going to be a huge question, right? Like is that your song or did you just prompt it? Did you say write me a song? That's really great. And then it did it to me. That's not a creative exercise, right? That's just the AI wrote something

What advice would you give as you write code?

I would find someone we still need people it's great to be able to do things solo

But people there's nowhere around that at some point here soon. I've been looking for co-founders already

But find someone who's savvy and something who's done it who built application But you wouldn't recommend doing it with AI on his own? I I think what you need to do is pull the reins back because chat GBT is like a PhD grad that just like had his first Adderall

use it to just fork over their data and access an agency.

I mean, nowadays, it's really more difficult for us.

So I'm going to answer your question with another question. Do you believe that knowledge is fine? Then there's always going to be the burden.

Would Google got in trouble for scamming all the books? Then this is the bigger thing is through this whole AI wave where they're like, oh my goodness, this AI thought of a way to blackmail this person or break its way out. It's like, we fed it Terminator, right? We fed it every weird subreddit that's existing of this doom scenario.

Of course that's in there. Of course it's going to think of the way to do this thing or self-replicate itself. Because we all had those ideas. All these great sci-fi novels have all the ways that AI could absolutely destroy us. And we fed it in there. It's not some magical thing of AI thinking by itself now.

And it's like trying to self-preserve. No, it read that. And it's just redoing. I would argue, too, are books just facts? A book is not just a list of facts that can be learned by an AI model. It's a narrative presentation of a lot of information and ideas presented in the right order and the right explanation with anecdotes and ideas.

You can read the blink of a book that summarizes it. But you don't remember the book very well because you don't have the anecdotes and the story. And you don't remember when you read it because it was just a summary. So I would argue that the facts that are in the book might be attainable in an easier format.

But reading a book is a very different thing. Especially, you're also leaving out fiction, right? Okay, fiction is for entertainment. It doesn't really matter if that information is in an AI model, right? It's about the experience of reading the book. And I would argue that why do we have paintings when you can just take a photograph, right? It didn't make paintings obsolete.

It's a completely different thing, right? So to me, a book is still a book. You just might not need it.

Human expression will always be. So on the personal level,

are you guys using AI for agents in your personal life? Or like Carp and Pease, like second brain?

I'm not. I mean, I ask it questions, but I don't have an AI friend or an AI consultant.

Like Sol or like Co-Work. No, I don't have something that's like, I talk to all day. I just, I give it specific tasks and I might ask a question. Instead of going to Google, I might go to ask Claude. But I don't have an AI, an agent that I generated that's like Buddy or whatever, my consultant.

That's a great question, Mike. And I've gone to the extreme where my iPhone right now is actively recording and filtering out and finding my voice and bringing it in, as well as taking all of my location data. Every single sensor that iPhone has is getting sent home right now to me through Tailscale.

And it's getting processed to see what I can get out of that signal that will actually become valuable, right? So him talking about like, there's no agentic company yet. That's actually the exact problem that I'm working on solving right now with that Adam repo being a free open source way to build anything you want, with Arthor being the way that how people find it, right? And then I have Eli, which is the one that's on my iPhone that's fully private, local, and it goes and it stays in the private cloud, right?

Where that's never gonna go to a model provider at all. And building that more as like almost a kernel of persistent intelligence instead of me prompting it all. It's sitting there and it's listening and it's thinking and it's building its memory store and making decisions there. And I'm like, these are the benchmarks and I'm publishing this all on my website and just building it in public to like get help from other people that come and look at it from different perspectives.

But yeah, that's exactly where I'm at. Have you looked at any frameworks or like you said you're building on yourself but like there's also a lot of like Hermes and OpenClauds and some of those. Yeah, absolutely. And I think where you would actually like to go past that would be Hyperion, right?

And OpenClaud because they've been working on this problem for more than 20 years. So if you wanna get to the next level cause like first it was really fancy about RAG and now everybody's really fancy about graphs and then next it's gonna be hyper graphs, right? And so, and it's gonna be a combination of those three together because there's three different memory types.

And so combining those together like a Hermes agent, right? And I'm taking from all of these open source projects I pulled them in and then I'm auditing through them to grab and rip out the parts that I want. And then not even taking that code directly into my code base, but I'm taking it and then making my own version based off of it, right?

For the good patterns, the good examples like the good thought process that went into it and pulling it and making it for myself. No agents for me. Last question here. I have a bunch of questions. All right, last questions. So you can ask this unsaid question for a reason for that.

So how different is vibe coding to agentic coding?

The great, so that was kind of what we talked on the first and he corrected me a little bit on the proper definition of what vibe coding is, right? But you were talking about like agentic engineering versus vibe coding. I feel a lot of agents will know like general has recently released Spark data. Yeah.

Spark is great. The agent's voice is standing on you and handling the similar responses

that is going to expose to vibe coding and appreciate between the agents

and using the coding code whereas I see things with their exception. There isn't really a distinct, to me when you are vibe coding, you are doing agentic coding in one format, right? You're just having it generate code. Whereas working with an agent, the agent can do a lot more than write code for you.

If it's responding to your emails, to me that's not vibe coding, that's generative AI in response to some trigger, right? But it's not writing code, right? It's writing a response, it's generative AI. But when I sit and write code, I think of it as more like agentic engineering where I'm telling it very specifically what I want, I'm looking at what it does, I'm correcting it, I'm tweaking very specific layout, design, architecture.

So it's like a collaboration where it's doing a lot of the laborious work and I'm doing more of the direction. So to me, it's still a form of vibe coding because I'm not handwriting all the code anymore. But it's not pure vibe coding as you described, which is just like, I, here you go, you do all the code, I don't know anything about that.

Like that's the more pure form of it. Yeah, 100%. But to me, agentic engineering is more of a type of engineering and it can be in a vibe coding way or it can be in a very collaborative, technical way. It depends how you do it. It's your discipline. Is it in the talking the way I do the work?

So people are like, do you see it? Do you use a specific language or do you think

it's a lot of language? So I've been more specific because first of all, I have like a large app that I work on that existed years before we started doing this, right? Like I was working on it for four years before I started using AI. So in that case, I'm like, look, it's TypeScript and React and CSS and it's got these libraries continue along that path and keep using this code.

But when I've built little projects of my own, like I've built a few apps in Swift and I don't write Swift code, you know, I can read it. I understand the basics, but I'm not an expert in Swift. But when I'm writing a native app for macOS or iOS, I'm like, all right, go ahead and write it in Swift. I don't care.

Because in that case, I'm letting it decide and I want the purest, highest performance tool. And I can still go back and look at it if I need to. But in that case, I'm kind of asking it. That's why I said before, like don't always tell it. Use this, this, this, and this. If you're married to that stuff, great.

But if you're not absolutely married to AWS, for example, or if you're not addicted to React code or TypeScript specifically, give it more of a general salute. It'll be like, this is the problem I want to solve. What is the best way to achieve that? And tell it, you're going to be writing the code.

Because one thing I don't know if anybody else has noticed, but a lot of times you ask AI, how long is it going to take me to build this, this, and this? And it'll say, oh, two weeks of original coding and then another week. And I'm like, oh, I'm not doing anything. You're doing it. And then it goes, oh, okay, I can build that right now.

Five minutes. So it's telling you how long it would take you to build it based on what it knows about you. But it forgets that it's the one that's going to write it. So you have to remind it, no, no, no, you're building it. So that's where it helps to remind it of that. But I think if you're opinionated about what language or what tool makes sense, tell it so. But in other cases, ask it more of the general.

This is one thing I experienced about as an engineer, is you'd go in a meeting and somebody who's not an engineer would say, can you get this API to pull this number and paste it here? And you'd say, can you tell me what you're trying to solve? And then I can recommend a solution rather than you telling me like, can you do this exact specific thing? Because to an engineer, it's sometimes like, well, why am I doing that? Is that the right thing to do or is that not the right thing to do?

You know, you're asking me something really hyper specific instead of asking me to solve a problem. And then me as the engineer, I can try to think of the best way to solve that problem. So sometimes you have to think of the, give the AI a little more credit and say, what do you recommend as the best place to host this that would be free and secure?

And you know, you would know how to work with it. Because if it's gonna be doing all the code, you might as well let it do it in the way that it's gonna do best, right? Yeah, that's great. And that was what I was gonna bring up is like, it was trained on the open source repos that are on GitHub.

And so if you gotta look at GitHub as it's composed by the different languages, like a lot of Python, a lot of JavaScript, a lot of React, and like microservice level, like that gets very deep and that gets to a production level for the MVPs and stuff where you're really starting at.

Highly recommend using Next and TypeScript for the backend. And it's a single application, it's full stack like together instead of having to deal with multiple repos or do any of that other mess. It's an extremely fast way to get your MVP out and into the world.

You said that right now you're doing a lot of prompting with the AI so it doesn't work for you. So right now, that wouldn't work for the AI for review like you said, for you isn't the easiest for the product. So sometimes I do have like, I even do what's called a, you know, an adversarial review.

Where you say, don't just look at this code, look at it like you're trying to make this person look bad, you know, like look at this and try to embarrass whoever wrote it, is essentially, you know, I'm exaggerating something, but you're saying like, don't be generous and say, oh, that's probably fine.

Tell me if your job depended on this code being good, what would you criticize? And sometimes it's better to have a different model do that. If you tell the same exact model in the same code session, hey, go review the code you just wrote, it's gonna say, hey, looks great. But if you have chat GPT review clawed for you, then you'll find that it'll point out a lot of things that make a lot of sense.

And that's another layer of like adding a little bit more checking, don't just trust it implicitly. Because I have like an auto agent that runs whenever a pull request goes up and it looks at everything and says, hey, this AI suggested this, but that's gonna catch these other errors that you actually don't wanna catch and it'll find a lot of useful things.

That's great. Yeah, so I use cursor, so it has a built-in bugfog, which is a separate one, sub-agent that runs specifically for finding those errors. I also do the same thing with an adversarial review. So most of the things you've been saying to that, I'm just like, oh, this is so awesome.

So yeah, that same thing, and then using different models to look at it as well, that's great. It also helps if you're an engineer to look at it yourself too. Right, right.

So this is a big thing, right? The needle in the haystack. As that context gets larger, its ability to find which pieces are the most important degrades, right? And so for mine and the hooks and rules that I have set up inside a cursor, it never goes past 60% of a context window. And I've built commands called the handoff where it'll take its context that it has stored in its memory, where it's at, all the things that are going, and builds that into an MD file that then seeds the conversation of the next agent.

So you can either tell it, like a cloud code, I just say spawn a new chip. It pops up a little, what it needs onto that new prompt and that can start a new session.

Or you can do it with an MD file, or you can say, write me a prompt that I can paste into another chat session. Or you can do basically a memory compression.

You can basically compact the current session that you're in. You can give it a command in cloud code that basically says, hey, compress everything back down and remember the relevant points. Because a lot of that context window is filled with stuff that's not necessary, right? You told it to read a PDF and then tell you the important bits and it's remembered everything unimportant from that PDF.

So you can use commands like that if you want to keep within the same session. But now and then it's worth just kicking off another one. And usually you're not doing enough work on the exact same problem anyway. It's probably good habits to just start, I start new sessions for every new idea, every new feature, every new bug.

That way they're all named nicely. They're organized well. You can archive the ones you're done with. You can say, commit that code and then go bye-bye. And it's a lot easier to organize that. Because if you end up with one magic session that has like 15 unrelated things, it's really hard to scroll back through that and figure out what it's doing. So at the same time you're confused, it's even more confused than you are. And I highly recommend going to a new chat rather than doing the chat compression, right?

The context compression. Because yeah, it'll be able to get 90% of the core valuable things out of it. But you're gonna lose 10% that's gonna ultimately change that LLM's output or that agent's output off to a couple degrees, right? And so for somebody like him that's very experienced and has been doing this forever, he can catch that and bring it back. For a lot of us that are not the really well trained or experienced, that's gonna end up taking you in a direction where you're gonna end up having to do a rather big clean up, pull yourself back, like oh no, this went this way.

At least that was my experience. Let's give him a hand.

So let me just wrap up with some announcements. So this Thursday we're having a start-up summer party at the Urban Hive downtown.

It's actually a great party. So you're gonna have, there's actually a silent pitch conference.