from 2x to 10x

00:00we are recording okay what's up guys what's up guys oh my god i have the whole intro and everything what's up guys just two ex-engineers here my name is matt and i'm joined here by the one and only big bro dc what up what up gang 2x engineers how how that joke has only gotten funnier as the the months have gone on because when did we even start uh having that conversation because i think you blew up like in my dms one day months ago talking about you just hate the idea these like 2x 10x 100x engineers oh it was first

00:55the 10x engineering it was 10x engineering that's what it was okay i thought it was a hundred no it was first the 10x engineers because i was like quote unquote the industry standard right if you want to be good you got to be the 10x engineer yep but now now it's the 100x engineer that you have to be the yeah there was that one post right you shared that to me on linkedin the guy that was talking about we have 100x engineers yeah now you got to be the 100x engineer if you're not the 100x engineer you're not you're not doing it right yeah it's kind of crazy like what they're asking for the output for us i'm i'm definitely a part of the the mental health craze of the energy gums have you ever done those energy gum yeah like a neuro gum i haven't tried it which by the way i'm just saying neuro gum please sponsor us is it good i like does it work i think it does

01:57i think it's a little i think it's it has passed the placebo effect for me huh yeah i should try it i always saw it but i never tried it i was curious yeah and i recently got their competitor so they'll probably not want to sponsor us after i show the competitor but this one's called respawn very very simple huh you know simple simple labeling um it's yeah it's it's the maker of five so this is their oh really okay yeah they're their their neuro gum uh adjacent but apparently it's also in conjunction with razor like gaming peripheral razor let's see on the top right huh yeah interesting so interesting but i will say i definitely did feel a real effect yesterday yesterday two days ago it's like two or three days ago and i bought

02:53it i bought it randomly when i was in target yeah it has green tea and b3 b5 and b12 i see i see but you know we're not here today to talk about my my stupid products or what i'm doing health and fitness wise no no that that's a different podcast for a different set of folks or just me speaking into a camera with no one there uh we're here to talk about ai something that you and i have unironically uh bonded over the last few months um being two ex engineers ourselves in in this new landscape being able to do kind of anything see everyone do anything really literally see everyone do everything kind of yeah both scary and fascinating at the same time we're seeing a bunch of people that are able to take their crackpot ideas

03:52and just crap it out in in a phrase that i continue to hate every time i see it this thing i one shot man i just like one shot this thing with the latest model which just we can we can we can get into that later but i love the one shotters the but it it does feel empowering and i can't get mad at them for that if anything i am so happy for them for the builders that could never build for the dreamers that could only dream they have the power now oh yeah i think it's i think this new age is more so for example like the people that have the ideas have the thoughts yes right and if they're capable of understanding how to use

04:47ai and actually go through the process of going through roadblocks when they come they have now the power to build whatever idea that they have in their heads um previously that never existed either you either had to be an engineer or a designer that knows how to engineer that has some product sense so like those three i guess the trifecta really yeah basically the trifecta field had to work together to get a product out a software product out right but now today it's kind of like if you have the ideas as a product person or engineer now and you know how to engineer and you kind of know how to design if you have taste and basically design which we can get into taste later yeah taste is a whole another issue another topic but if you have any of those three things

05:45right you can basically have a one-man army one-man team to build anything you want as long as you're diligent in figuring out how to use it all yeah it's it's kind of kind of incredible just how much people have been able to build and how much it has even changed in the last six months six months six months because we were even just uh gagged is probably not the right word but at every installment of the latest opus and then when uh chat gpt released their latest and greatest model like apparently it's doing great for a lot of people which compared before chat gpt wasn't even in the discussion for like a good coding model yeah i think it was like a basic like oh it kind of gets this stuff

06:42done and people were kind of using it inside their cursor or ide or whatever like a bunch of engineers are still getting their ai credits from the companies just from like the simple you know 20 plans enterprise plans and they're just baking it into their ides and they're like oh yeah i used ai but like in reality if you want to compete with the ai frontier labs they're not just using your your 20 monthly credit or whatever no we we're the freaks doing way more than people can even understand right now because every time i'm pretty sure you've experienced this i've experienced this enough times that i can confidently say every time i sit down and i'm just like oh let me show you how how fast i can build something and i do it in front of their eyes plain english with what looks like

07:41a an orchestra of people and we just somehow build a symphony on the spot and i'm the conductor being like yes i don't know how this is happening but i'm doing it but it's great but it's great i think it's i think it's great in a sense like for specific features that we have the context for of how it should work it can one shot pretty quickly fairly easily but if you just give the ai like no like no guardrails or like some sort of guardrails yeah and you kind of let it go it could get like a pretty decent maybe working solution out there with an okay looking thing which is what most people that are not in those three correct i guess fields yeah are probably like oh wow and then like they might

08:42get turned off of like oh that's not exactly what i wanted but in reality like you didn't give it the context of everything that you need to do all those things so then bugs may come into place plenty of times it's not even just bugs it's there and this just comes down to one of the things that i love doing i love watching people's brains work i love to see there and i love to just stare into their brain and i watch the little cogs turn of like how do i convey something to someone else especially when it comes to like teaching anyone that teaches anyone anything i love to be the person watching the teacher so like for example when i watch people use ai for the first time or not even the first time i just

09:37watch them handle their prompts and like how they go through speaking to the ai to get their desired outputs for every prompt that they do for every string of words that they think of either say out loud or type into the field you are seeing firsthand how they are like constructing their thoughts and i'm like and i'm gonna have to be pejorative in what fucking world did you think that that sentence was gonna get you the output you wanted i that that thought comes across way more times than than than not at all and i can only help but one be in awe and then two have to now figure out the way to convey to them that you are speaking to an entity that doesn't get tired that will more often than not

10:35take what you're saying at like face value unless you've already given it the pre-established like context of it understanding your speaking structure through clod md or some other means of it understanding like your own tone between sessions mind you which a lot of people also aren't handling for themselves in such way to repeatedly like get yourself to the end goal and not run out of tokens which is also a thing that people experience which i need to know the stats like that's that's one thing that i i wish i had access to i think there's a lot of the people that are on the 20 plan it's really just the 20s not like the hundreds or 200s no it's primarily like the 20 plans people that are not willing to spend a hundred dollar

11:32on the mid mid-tier um and they're just trying to like a token max with 20 dollars yeah to me to me and they might be doing that with like two accounts so then they instead of spending a hundred dollars they're spending like 40 20 20 yeah 40 with clod 20 and then codex 20 so then they use both accounts that way yeah but like to me it's kind of like you're telling me you want to be on windows and on apple at the same time because they work a little bit differently in their environments like pretty similar but also if you want to like supercharge one environment it's like you're working in two different things yeah you might as well in my opinion spend the hundred dollars on just one environment and then make it extremely good and then your output is just from 2x to 10x

12:36how to go from 2x to a 10x engineer you can just go and pull up just we use one plan and get really fucking good at that one plan to get from 2x to 10x and then even with that one plan you can get really really really really really good and still probably token max to get to the 10 or 100x 100x engineer i mean we're definitely hitting that point where there's a new solution and or a new way of doing things i was gonna say every week but i feel like the i think it's slowing down a little bit it has slowed down i've actually noticed like a lot less has been getting pumped out it's because these fuckers don't even know like what they're talking about half the time it's like a new oh we're going from prompt engineering to context engineering from context engineering to what is it now harness

13:32engineering harness engineering to then the newest the newest one loop engineering and it's just like bro all that shit was already there it was there the whole time from the time of prompt engineering yeah so it was there the whole time and you guys are just thinking of like new ideas to like put your stamp on it and to make content around yeah this is the new loop engineering yeah like loops loops exist existed in engineering since the start of time and and even just to get slightly mad at loop engineering because we're entering we're entering july in t-minus two days open claw that was popping off in what january was that january no i think it was like the spring

14:30like probably like march it was a little no dude i think it was i'm being genuine search it up wasn't that fact check me when did open claw like explode in popularity i don't know i thought open claw made the pretty sure it's like hard happening i think open claw open clawed i think it was january i don't know they don't have a timeline let me ask ai when was open claw launch when was open claw november of last year interesting that's what i'm saying you're probably you're probably right then the the popularity started in january that yeah that's like that's the whole thing i'm i'm trying to harp on

15:29we have been in the world of making a looping system you're always on bot it's been there we've we've been there like to me it didn't make sense because then the for example they're always on bot right like you say just say nerd you could just have clod or you know codex on a machine just like install it on a machine and then have something run those terminal commands with clod code and then it's on like as long as you have a server or a computer that's on it's automatically on like you didn't even need open claw you you could have done that on clock code i mean i already had that because i wrote a server that basically listened to my linear tickets yep automatically and it'll just send like a linear ticket to your unit to to like i didn't even have

16:32it i didn't have it on a machine i had it on my local computer i just i just opened up a port and then i would just send it locally on my computer and then it'll work on tasks and i was just i was working through and like um what was it called having my clod just write linear tickets and it was just fire off fire off over and over again i like that yeah that's nice that's what i was doing i've i've gone away with that yeah because at the time i was having to go through those branches and all those tickets and then like just review the work and then it was just so much code output but i couldn't even validate and verify the code and i was just looking through the code reviews and i was just like dude this is going to be the bottleneck i have to like review the code and half of it was not as what i wanted it might be like 60 there but not yeah 80 this was before

17:324.6 or yeah this was like 4.5 or something which is pre-november yeah yeah yeah so i mean that's where it was previously of like i think it was like 60 almost 70 which is kind of true now it's 4.8 is like you can get 8 pretty easily yeah i don't i'm getting far less mistakes now yeah i'm getting basically there but then the context of what i wanted in terms of like design was like misinterpreted either misinterpreted i get the and or and or wasn't um pushed to actually like define of like this is what should happen and claude had to kind of decide what happens oh and or like missed cases and or yeah and or it's like oh it will do this but then when you do it it's not actually working so it's kind

18:34of like a little buggy but not too buggy i mean you kind of just have to work with it massage it a little bit and then it'll get there yeah i can definitely understand that especially working on the automations portion for big because i really did i had it one shot a lot of parts of the design because i wanted to just focus on the technical aspect about just how the triggers occurred and then i came through it's like i don't like this we're going to tweak this now and i find that i like this approach way more what you mean i've in terms of my loop engineering oh i like to give almost what feels like a mountain of context up front around the i guess unironically like the user stories user should be able to do this when user interacts with this this should occur like

19:32i try to list out all of those and then it when it starts to ask me like design tickets i give it like a super vague idea unless i have like an exact ux that i want because i just want to make sure that those user stories and all of the underpinnings about handles the data is like done then it's like cool we have all those functions we have all the manipulations now let's figure out how the ui does its thing like the the quotes hard part is done now we can like figure out how this thing should actually function my my my way of loop engineering is a little different because you come from like uh design aspect yeah like i want i wanted to look a certain way but i'm like on the back end side of like most of the stuff where i focus more on the data set first and then i work my way up so then it's kind of like yeah i think yeah i definitely i think user first yeah i don't think i do you think about

20:30user but i make sure that the data and everything like that makes sense yeah it's more sound first and then like okay with this game data set does it solve like this user story or user journey right but most of the time it's kind of like working together with all those things so you could kind of just say like okay here's the user journey and then like here's what i'm thinking about the data does that solve it and then how and then i i asked claude like with that give me some options of how the data should look like right how do we store it what's actually needed for the current iteration with xyz users with xyz expectations right that's that's a that's a key thing right of like expectations of the usage yes of the actual huge is huge because if if it's like a big user base and you tell it that it's going to design

21:32something definitely for a larger data set right yeah and if you don't have that many users and you're just trying to get a feature out as soon as possible they'll also do that right but most of the time i think it's going to default to the to the later one where like it's going to assume that you don't have that many users yep i think i think that's the case yeah i think that's you it's that's the whole one shot yeah the one the one shot one kill pros out there your your ai agency builders that aren't engineering background yeah those guys will probably have that type of uh aspect where they're building these businesses or features or solutions software solutions for businesses and they don't even know the context themselves yeah but in a lot of ways i think even that is being

22:32commoditized and that's actually one agency uh the people trying to one shot and making sure all of their ducks are in a row what i mean by that is you know having the security side making sure the data is like handles xyz what i mean by this is we're we're even you and i are already passed part of it to a very soft degree depending on how you use it with the paradigm and its agents because there are specific agents that do know about these checks and balances so when i go through my feature discussions with the team i very much outlay like the full product and then we go through it a number of iterations where it's just like this amount of people is going to need to touch this it's like this agent kicks in so i was like oh i need to make sure that this is designed it's like they're they're already they're pre-prepped ready to handle this yeah but then to to give context to like people for us it is very

23:32important because of the way that you've designed paradigm to like front load a bunch of the agents with best practices and everything like that and it's their one thing to know that's all they care about the agents know like the agents are like boosted to be senior agents right like people that are just using cloud code out of the box and using skills and everything like that these cloud code is not gonna or clod is not gonna be their out of the box senior it can be no yeah but what but you have to give it the context to be to for it to be that right like these paradigm agents underneath the hood is clod but we gave all the context and best practices for it to use in order to be that senior type of agent yeah and i think that's a one thing that i know you wanted to talk about was the idea about how

24:32how how users sorry what steps new people to the ai agentic building world should learn like how to use it and it's not that you're just like trying to build the thing you're also trying to unironically build the team yeah i think it's more so you're building the team and also you need to understand a way of thinking yeah yourself yeah because people are coming into it and they're just like oh let me just say like build this build that build this look at this skill look at this you know um mcp server or whatever and just connect everything and be like all right let's let's just build the thing and one shot stuff right that's how people are doing it no no no no no no that's not how you should start right it's kind of like you have to build the foundations of one how to prompt because when you learn how to prompt that's

25:33basically teaching yourself and cloud or your other model how to build context right and then from there is kind of like you go from there to then skills which is just basically a list or a folder of context that your model is going to reference over and over again and it has pointers or whatever right and has maybe some files maybe some code files that i could use here and there and then from there it's like okay how do we bring all these skills which is basically a bunch of context and placing it into like a more agentic version which is basically paradigm in a sense right but i don't think we use particularly called skills which with skills is just basically markdown files that anyone knows like it's just it's just it's just it's just documents literally it's just folders and documents that's all

26:34it is right anyone anyone that's like pitching like oh god build you an agent and stuff like that it's gonna be like self-learning and blah blah blah blah okay so how is it actually learning oh we just need to tweak it and change the context so you're just giving it more markdown files and and you don't need an engineer or or someone very technical to be like yeah we're self-training and all this other stuff like that's a bunch of bs unless they're building actual ai models or lms in the very beginning stages yeah for training like when they say training data a lot of these people are like oh yeah we're just training it what is your way of training uh we're just providing more better markdown files that's how you could tell

27:32who's the who's the fakers and who's the legit people that actually know their shit yeah so how do you really feel about working with ai agencies are they are they actually worth it to the to the average average businessman or actually maybe it comes in two fronts versus uh smb versus like large right no even even so like agency wise in any aspect i've always hated the reason why is because most agencies they don't have the context of your business you have to provide all that and then they have to understand how you do things if you are good and particular of how you want to do things like when you're in a big company you have certain ways of doing things for your teams or whatever and the agency the agency people coming into your team to work on a specific thing they might not even know your

28:29conventions or context or anything like that so it's kind of like you're training them for like a couple months and then after those couple months they're gone so you basically train somebody for no reason for like a couple months and then you lose all value and then you and then oh they were cheap though let's keep doing that over and over again don't waste my freaking time if i'm you know if i'm the person that's on a team that has to bring in people i'd rather just train somebody and have them on the team consistently versus training someone and then like a couple months later or a year later they're gone like to me i would never want to work with an agency because i worked with plenty of ngs agencies in startups in mid-sized in large size and every size and for whatever agencies even in the

29:28startup world they're like top rated right okay every person on in that agency has churned so maybe the top rated people in that agency aren't working with you so yeah i can't i can't say i've also had the greatest experience with it and that definitely comes at a huge judgment i've found to a lot of companies longer term but i think at the same time now because before software at least was extremely hard to get out so it's like you almost need an agency be like here's six people they're here for you they'll build for you they've done it with our conventions boom yeah like the most well-known company that i know that had an agency model was disney if you look at oh yeah if you

30:31look at any disney product their disneyland app their disney website the disney plus not disney plus because they tried to do that in-house right so anything that disney did previously with apps websites or whatever was all agency based and they're all frankenstein monsters because none of them work with each other and then and then they wanted to do disney plus and then they're like oh we need to actually hire good engineers and and they were hiring at such a low bar of like what was it called uh of uh compensation that none of no one that was actually good and wanted to work there wanted to work there until they started raising and being like uh competitive in their uh compensation

31:26right i didn't and then now people now people are actually there and also because of hulu like with the hulu acquisition and everything like that and where they are right now they're getting there right but then like any of those websites and everything like that is a total mess like if if you asked me to work for disney i would not want to because depending on what team you're on it's going to be a mess i really do wonder how we'll call it the old guard they're not really old guard but we'll just say for the sake of this argument old guard versus new guard is adopting to this ai world now because you don't need that number of people anymore you really don't and i've been telling this to a lot of people it's not ai is not replacing every job it's just doing an 80 percent reduction global yeah i think it's doing an 80 percent reduction but then what happens is for example those people that

32:27don't have roles there's either going to be a bunch of new startups and new businesses that just want to probably hire these people which i think that's kind of what happening like if you're ever in the i don't know industry right and you're looking for a role most of the time the easier it's always easier to find a role in the startup world in my opinion because they just need the people yeah versus like if you're going if you're if you're going to a large corporation they've got thousands of applications for one role like why are you competing and doing that but and also like if their compensation that much better i guess if you want to cruise if you want cruise control go to a large corporation if you're young and sprung go to go to a startup you're young and spry willing willing to sleep under a desk

33:27i mean not not if you're out of college i mean out of college like you might as well go for it right like also one the compensation is can be very competitive and also you might you might hit a unicorn and so you might not you get a great story oh you get a great story and you most likely you get more experience at your startup than any large corporation the great corporation will have your team and your specific little tiny aspect of what that team does and what you're going to work on that's it you're not going to cross-pollinate in all the different types of stuff right like whenever i go to like a mid-sized company and i'm like a specific back-end engineer and when any anything else comes into place it always goes to me they're like oh who can we who can we push to work on this oh give it

34:28to darren has darren has experience in that i have experience of that i have experience of that i could do that or if we work with another agency that i have to work with their agent not an agency and or another company and we want to integrate with their product yeah and they don't even the people at that company doesn't know how their product works i figure out how their product works and i teach the person at the other company that we're trying to integrate with how their product works i don't want to get into that because i was trying to integrate with a devops team with a specific company and then i was like asking them questions in their slack and then they were saying oh let me let us find who worked on that it was like taking me like days to find the person to get back to me of like who built the thing and then i was just like can you just is this open source can you just give

35:25me the github repo to figure out how this freaking thing works and i was like written in python i don't not even a python engineer right i just looked at the thing and i figured out how it works yeah and then i'm not a devops engineer either but i was basically helping in the migration of all our services from one platform to another and to make sure that i had like the right cicd pipelines and i was no one else in my company knew how to do it we didn't have any devops engineers i was a backend engineer and then i had to figure it out because i figured it out and then everybody else's projects or whatever systems just basically copied what i did because nobody else knew how to freaking do it and then i already figured it out working with the other company's engineer but i basically told them how to do it

36:20amazing man it's just it's just a lot of fun it's it's it's fun we get to be scrappy and build from from that it's also fun like being the guy i don't want to i don't like being the guy unless i'm in charge because i don't i'm not a one i don't get to choose what i get to work on if i'm working for another company and if it actually moves the needle one i could say and and have buy-in which i like if i have like a specific manager and they give me buy-in of what i want to work on but not every company does that and not every team allows you it's really sad when when in i was going to use a word that probably isn't even correct in a concordance of people there's a word in a room yeah that's a big word

37:19that's probably not even right a future matt place the place the the definition i was just thinking about how even in a room of a lot of minds in a startup i should use the startup right and everyone in the room can kind of agree on the idea but then there are select few that by just the nature of power dynamics and companies it doesn't matter how good the idea is we're doing that idea oh yeah they yeah sometimes sometimes it's like that yeah and you got to just be like okay we will make the plan do that yeah totally that's let's pivot again that's that's the reason why i'm not a friend engineer because there's a lot of small things like that in those type of power dynamics

38:15or if you're an employee you kind of just have to say okay sure yeah even if the idea or thought doesn't really move the needle or is important but nowadays that's kind of to me that's kind of gone in like those simple small little changes if that's the case it to me that can just come from the product and or designer immediately yeah versus going to the actual engineer that needs to focus on like more thoughtful uh features and or ways of getting the product out that make a dent which is also very interesting if you haven't seen that like um anthropic and open ai and those people like the product teams and the design teams and any like super ai focused or forward teams will have their designers and product

39:12like people actually create prs we're definitely hitting there and yeah we we at least know that we're not hiring a designer unless they have clawed open running prs and making changes they're just yeah hired no i think so too like to me did i tell you about my my my tune-in exam interview that i that i want to do no i haven't seen let's see the tuning exam so i've had this idea because it only made sense with regards to the new ai wave but in a world where you know we have the company it's up and running ready to do a wave of hires right we're looking for two engineers two product two designers right we're just hoping for essentially two new pods each pod being one of

40:09those triangles which i think is like the perfect balance right then you know if he's sick one day you get like 30 or 60 of the brightest young minds that want to work for for uh for work for big they all come up it's like oh what did you get in look i'm in the designer brackets like okay cool cool cool and then as soon as you guys walk in you get split into your tables and you have to go through a series of like one knowledge challenges like how much of your discipline do you know and then some people just might be cut out right there right then next test actually no you come in no teams yet revise come in no teams you do your test your practical like simple test about do you actually know your stuff like pretty if you're really dumb you're just not passing right next you get grouped into random teams and then maybe for the sake of the tuning exam

41:04where you're like building a product there'd be like a literal practical where you have to pitch an idea and build an idea within like within four hours maybe like maybe six hours right it's like a full day affair and there's like every two hours or something there's like a break where the prompter whoever's running the interview gets to throw a wrench it's like he goes to every product and be like you know i want you to change this about the product i don't think we want to hit this market anymore i want to hit this and then like watching how they speak how they fight back their idea and like you get to see so many different things and like how well they work one as a team and in pressure as well as like also being able to see like how like what kind of product or change on a potential existing product of big comes out and then at the end we choose we choose

41:59the literal best working team it's like i can see us working with those three people or it's like we want you two and that person yeah i think it i think in the new age it's more so i've been seeing this a a little bit since the start of last year where some companies will try to actually get you in for like a good month of working with them yep and kind of vetting like they gotta see how you actually vibe yeah how you actually work and how you actually vibe because in now nowadays at today's age it's kind of like if you're interviewing on their knowledge and how much they know i don't think that's the actual test no i don't really care how much i don't care how much somebody knows like it's great if you know something but i care more so about if you don't know the knowledge how can you find it yes and how can you distinguish if the

42:58if the information is correct or not and how can you apply that information yeah and so application vetting and adaptation yes i think those are the three things that i want hiring for i'm not hiring for how well you know something ai knows like the triple a percent of the stuff the triple a's how well they can assess apply what was the last one i don't remember adapt adapt access apply adapt can they do all three oh also are you this kind of guy this kind of guy or this kind of guy you do you do the reverse three so when you count to three you do you do this you do three when i go

43:54three i say okay i don't go three i sometimes go if i count i might do like one two three yeah i just want to throw the three i might just go like that yeah this buddy oh that's cool because i was just like i was just like one two three i did i did thumb uh index middle i was just like oh i did three like this but if i said three i'd probably go three like this but then three like this is also great yeah anyway yeah i think that's how a company should hire because let's say you know all the knowledge you know all the jargon you know everything like that if you can't apply it who the hell cares yeah right and then i guess for engineering aspect it's like do i care if you can code not really because i

44:52don't even write code myself anymore no but you're able to like sniff it out yeah i only care if you can one read it and also get ai to do the work that you need it to do that is all i care about sure same thing with like any aspect or any piece of work it's more so like do you know how to get ai to do that piece of work at the level of expectation and a high level yeah true and do you have the criteria and or knowledge base of what is considered a high criteria i think that's maybe the knowledge that you need to have i think and or to find that knowledge you you brought us all the way back to one of the first points you made which was taste actually being able to say this is a good steak

45:48how do you know how do you know how do you know is this chicken cooked or is it wrong yeah exactly right even so you could get like ai to give you like recommendations and options but with those recommendations and options how do you how do you how do you vet between them right how do you know which one actually sounds right that is i think what comes with knowledge and experience that you after all looking at all these different options which one makes sense and you having that judgment right i don't think ai can make that final judgment call i think a human person can do that and or at some point in time to actually have the correct quote unquote loop engineering you are teaching the ai your judgment that is actually loop engineering and getting it to actually

46:51that's the final piece it's not like oh can you put this on a loop and whatever it's like can you make sure that the ai has the similar judgment call that you have when it is at a higher high pace level yeah it's uh it's a self-correcting you know agile sprint loop we're just we're just doing that forever because if the loop is not getting faster you know you're not you're not doing your retrospectives correctly i dude unironically i do a lot of retrospectives with my with my agent team do you i do a lot of them i i do retrospects only if i have problems if i have no problems i may do it that is dude you have to do it for both otherwise it doesn't know how to reinforce well i will make the reinforcements when i'm at once i have a lot of down time i will go back and

47:52i'll be like okay i found all these problems well i usually it's always i find problems to make sure that it has like definite like action points to put into itself to make sure that none of these things happen so you're you're more along the lines of really focusing on damage mitigation for the future yeah because if it's already good and it already gets like what i want out of it how much gooder can you get oh dude it's so good it's so good how much good can you get oh see it's already pretty freaking good but it can get it could get gooder it can get gooder but like at that point once we figure out the judgment call pieces like if we have a judge the judging agent that makes the final calls that actually

48:49knows yeah i mean sometimes like sid sid and or the agent isn't actually doing the right judgment and i'm like no and then you gotta teach sid and or the agent to be like better or gooder yeah um which is which is the retrospective it's definitely the retrospective but like also you need to lower more you need to lower a lot more i have i think something's going on with like how it's not pushing the lower i think i have a lot of the lower local oh okay yeah tell your agents to to always push uh local lower up to the branches otherwise the stream is you know it's devoid of all of the the pieces of the timeline history that's like half the reason why i'm i'm trying to like my entire new source control is really getting its bearing right now because i'm basing it off

49:45what i've learned from lore i think i have lower well yeah my my prs have lower but maybe i'm not like i i know like sometimes start certain you know well think of it like this way anytime there's a major junction or a big decision or what i like to do a lot now is before a major plan occurs i write lore and i write lore specifically to track how the agent team believes believes that their solution is correct i want them to write it down because when it ends up when it ends up finishing they get to write was i wrong and why was i wrong or was i right and why was i right etc a lot of the heavy big players that i know that actually makes sense we're all talking about the same thing of like when they integrate ai into their business how do you have ai have the context of how things run yeah in my

50:47opinion how we do it is the lore oh like lore is yeah i love yeah lore's one of my favorite things ever lore is the context that the agents need to know how things are working and how things the decisions right like if you're trying to build an ai first business where agents are across every department and that knows how everything runs and how everything was decided and everything like that that's basically lore yeah every decision of every thought that an ai agent makes is not only tied in lore documents but each lore document specifically annotates to specific commits oh yeah so there's a lesson that's specifically our engineering aspect yeah and design like basically any software product company that does product right in the software sense and startup in that sense yeah

51:45that's kind of like the a solution of how to do things like a self-documenting system of however they're documenting their decision making or the agent's decision making to then quote unquote train better their agents or anybody else coming into the system to instantly access the lore and be like okay give me a timeline of how everything was made and it could be just literally like how did automations start on big how did everything start and how did it get to where it is today yeah that's the floor that's the story that's some that's some fucking lord dude that's that's the story of anything do you do if you want to rear a weird uh uh time capsule do have your agent give you a summary of how the fuck paradigm turned out the way it did

52:42look at that lore because it has gone through some changes yeah yeah that's me and like a lot of different projects but that didn't have a war and i'm like man okay you wish you had this story when i had when i find things emotional like oh i found a new like library or technology thing is it actually useful do i actually wait a minute i should just use this i should have just used this in the first place oh shoot now i gotta fucking migrate everything oh great here we go here we go fuck all right my adhd brain is adhd-ing it's definitely doing it yeah yeah making making those huge pivot changes is it's nostalgic at this point oh my god i used to get so mad at it but nowadays i don't because i'm just like

53:41oh let me just click click a new instance or uh you know a new terminal instance and uh enter capy tap enter baby and you hear a little you hear a little bloop bloop on the computer ah yes it's ready no no no it's the enter button sorry oh yeah i had to hit accept for you to continue yes sir no that's why you have a bluetooth you need you need something bluetooth for that aspect pro tip for anyone that's using anything like that you need you need a actionable bluetooth thing for your computer you know i've heard some people have something on their glasses yeah yeah some people have like a button right mine's just a bluetooth the keyboard i just you know i need to use my my voice solution more because the day that i did it was

54:38pretty it felt it felt wrong it felt so wrong it is it's tony stark baby like it's yeah it's time we're getting to that we're getting to that just wait until the robots come i mean they're here like 30 grand they're there but i don't think i don't oh yeah like one more year they'll be solid they're not here in the us because they're not well they are but they're like in beta and they're not releasing it to everybody right yeah but in china i think i think it's rolling out i think it's rolling out they got like fleets of it already it's over man my robot is coming my robot is coming i'll have i'll have that sweet laundry folder of mine or that all right you probably can't do dishwasher because the dishwasher you know uh i mean they already do they could they could pack the dishwasher right yeah no

55:40but i think i think a lot of them they have like extendable arms so like it's like the the mechanics of i've seen them they're like hooks or whatever so they're like extendable hooks and they can like still grasp and manipulate things it's fine we're just gotta accept that they're gonna be here robot dog is the future i guess yeah i haven't seen that if i had a robot dog though that would be lit i would have a lot of fun with that dog i have a little robot guy here that is it is it a roomba it's just no it's still in the box oh no it's like this elk let me show you guys i haven't opened it it was a gift from my from my family i think last year for christmas yeah but it was like this little can you see it can you see it uh oh oh oh oh oh okay

56:37i think i've seen yeah i haven't i haven't i haven't tried him out he's done he's still unopened he's cute i will get there i know yeah i'll get there when i get there maybe if he any hardware machine that lets you download software i will gladly take over just wait man let me let me go through my uh raspberry pi era in like oh my god i'll say i'll say it's like my 2027 year it's the raspberry pirate pi era where i plan on just building up a bunch of things just a bunch of things uh yeah it's gonna be it's gonna be a fun world um so you you posted here a couple things that i wanted to talk about but i didn't fully know what they were all i see from you is that you said yc trash

57:38ah yc trash companies yeah basically like ai rappers oh they're like oh so companies that just do rappers around chat gpt etc okay yeah there's a bunch of those but then also it's kind of like um is also the product isn't that sound or like thought of and i don't know how they're becoming a yc company and it seems like yc and or these um funding companies right are just trying to pump the ai and see if like anyone hits but i think it's i think it's majority wave has changed yeah but the also like it's very interesting because sometimes these people don't even have a product and it's just an

58:39idea and then the idea is not that and it's not been right it's not been like a good idea and then the marketing might be okay right but then it's just like are these people actually going to be able to deliver that's first of all like a lot of these kids that are coming out right sometimes they don't even have a background in any like engineering department or product department and you're just like 23 year old maybe 24 year old with like ideas that are solving that are trying to solve an idea that who i see or whoever has created for them to solve and it's just a batch of them and then they're all going on the hype train on like twitter and everything like that and it's just like

59:36okay that's great and they sometimes get funding and then like after their first or second round is the funding's just gone and then they're like oh tough out of luck but then it's like i could have told you that like to me it's like why be more selective is what i want to say to like younger people be more selective in terms of like actually trying to build and learn something to actually do something seriously is what i would say maybe you take it maybe you take that idea seriously but then it's kind of like okay the taste of the product right is the product actually something that people will use is it actually thought of from the consumer side if it is consumer related if it's not a consumer product then

1:00:35whoever your user base is is it actually something that they want that's a whole another thing we've we've been in plenty of startups where the owners and people don't even care who the users are they just say like oh we're just gonna build this product it's like have you been asked or thought about like who's going to use this why would they use this oh yeah does it even make sense technology that target yeah like does that target audience actually want something like that and or is is something that is easily accessible for that target audience for them to use that's another thing it might be easy for them to use but if there's you know different regulations or guardrails or something like that that's another topic yeah regulations and guardrails that all these startups try to just bypass and

1:01:32they're like oh we'll just we'll get to it when we get to get to it when we actually have traction no then we'll then we'll find a legal team and then you just get smacked in the face yeah and then they're like and then all the traction is gone yeah and they get fucked i'm just fine matter of instant like right they're like oh we'll figure it out when we get there we got there bam goodbye yeah it's it's gonna be weird it's gonna be weird but at the same time i'm i'm i'm generally excited generally excited for the future i'm also genuinely terrified i have been a long time doomer of where i've thought things are going i've you know i've seen the tea leaves i've history repeats itself and the same the same game is going to continue running until

1:02:32until we're all dead but it it's fine we're here i'm thriving it's gonna be fun it's gonna be interesting it's already kind of fun but it's already not but it's okay we're we're here we're here and it's a good time we're having fun we're making big changes making big changes yes yes yes is there anything else that you wanted to uh because there were there were some other things in here that i thought about talking but i feel like that's a fun a fun rabbit hole we can do for episode two i think some there's some topics are more like in depth i feel like yeah well on the next episode i want to dig into a bunch of github repos that are shared and like how people say that this thing is valuable i want i want i want to dig into all everything around that yeah let's let's go look through like look through some

1:03:36like github star repos and be like yeah it's just a discussion about i don't i don't want to get into it now because it's like i've been mad for a long time but dude there's there's people that are just like oh yeah this is the latest like ranking of the github star repos that you should be paying attention to it's just like don't forget to comment repos are you on i don't want your trash list bro i would love to know also the stats of how many people one comment this list in terms of you and then three skip two skip no get to the point of how many people actually test the repo they don't that's what i'm saying but nothing is like these people don't even think about like

1:04:33one security issues of like they're just like well let's just use it let's just use the thing someone else is using it like dude like what npm got hacks and what act what axios one of the biggest like one of the biggest packs packages that's used globally biggest well-known packages used well-known globally with millions of downloads which by the way i was unaffected oh yeah apparently uh uh on my one of my repos way back when the thing happened we were like three four versions behind for npm uh yeah for for whatever that issue was we were just chilling on the old yeah but no because it was actually like a new release yeah it was specifically what is that type of attack called man not man in the middle is that i don't know i don't i don't remember it's kind of like

1:05:31that like these people that one they might not have like an engineering security background of like that's one thing right if you're a good engineer right and you're at a good startup or a good company a lot of the times the the head honchos the more senior people on your teams yeah are gonna look at these packages and they're gonna be like what package is that no we don't need that like that's not don't need that package no because they will they will look at it and see if they actually need it and or if it's make sure that it's like doing the right thing legitly i've looked at plenty of like packages and either taking like a snippet of the code piece that i actually needed versus actually downloading the full package because i don't yeah because i don't want to update the full package and then like package manage everything like that and then no no no no no forget that let's

1:06:29just take the take the code take the small little thing put it in a little library class or a library file and then just use it throughout the code base that's all a package is really you just get version updates but you don't you know you're relying on some third-party person that's outside of your team to update it in order to keep it secure let's not do that that's gonna be fun we're gonna do some show and tell because there's some specific packages and things that i want to talk about and you'll probably have some like fun ones that's and also the same thing with skills like clawed and skills it's the same idea like people have no idea and they're just like oh let's just download this skill that was good and you have a list of freaking skills it's like clawed and it's just like a bunch of freaking noise and a bunch of junk sometimes and it's just like these are not even useful but the stars yeah but they're cool

1:07:25and or like some big company released a bunch of skills and it's like what is this actually for does it even actually make sense to use any of this after the time no but it's but a big company released it it must be good right it must be good it'll just make my team smarter somehow yeah yeah i don't know how but anyway it's been great chatting with you darren we are in fact 2x engineers trying to build something big we should do that that should be like our 2x that's that is how the episode ends right there oh my god i'd be so cringe but so so