{
  "video_id": "mapbeJh5g98",
  "channel_slug": "dataindependent",
  "channel_handle": "dataindependent",
  "title": "She Makes 6-Figures PER PROJECT With AI | Founder Interview",
  "duration_seconds": 2615.0,
  "url": "https://www.youtube.com/watch?v=mapbeJh5g98",
  "upload_date": "",
  "transcript": "will deliver like an entire application\nbuild in like from scratch in 4 weeks\nlike fully functional we're not afraid\nof anything we try to take on the\nhardest projects we believe that access\nto hard problems is like a proprietary\nbusiness value for us I'm taking like\nswars of code from Claude projects when\nI'm working on client projects what I\nlove love love about Claud projects is\nthat you can share them amongst your\nteam one engineer told me he's like you\ntrust it much much more than I would\never or trust\n[Laughter]\n[Music]\nit that's Nicole founder and CEO of Head\nStart and applied AI Services business\nshe started the company right at the\nlaunch of chat GPT and now she's\ncharging a minimum of six figures per\nproject with only four employees in this\ninterview Nicole shares the Homegrown\nautomations that keep their revenue her\nemployee High how she uses Claud\nprojects to run her entire business and\nthe Playbook that she's using to scale\nher business from three to 14 employees\nlet's jump into it so Nicole you and I\nfirst got introduced by a mutual friend\nand he says Greg i'm met this person\nthey just started their own agency and\nthey're absolutely crushing it can you\ntell me more about what you have going\non at headart and what y'all are doing\nabsolutely so I started the company uh\njust over two years now I left my job\nOctober of 2022 I actually created the\nbusiness in December and between that\nchat gbt launched to everyone uh\nNovember 1st of 2022 so it was quite\nlucky timing in all of it I had actually\nhad early access to the open AI models\nand so I had been using gbd3 before it\nwas a thing and I knew I wanted to start\nmy own business but I didn't know how to\ngo about it and so I started Consulting\ncoding in line with the chat GPT launch\nso I was using chat gbt 3.5 to write\ncode for me before everyone was doing it\nand I remember to this day people being\nlike Oh chat GPT 3.5 it's not good at\nwriting code and I had a Consulting\nproject that I was on that was Ruby on\nRails I had never done Ruby before and\nso it was writing all my Ruby code and\nI'm like well it's good at writing Ruby\ncode so I don't know if it's something\nabout the way that I'm using it or\nwhat's different and then by the time\ngd4 launched in April of 2023\neverything had changed and so obviously\nsince then we've gotten almost weekly\nupdates from anthropic and open AI\ndropping things that use to code and so\nmy business Head Start is an AI native\napplied AI Services firm and so it's\nvery meta in terms of like how we use AI\nso we use AI to write code that\nimplements the AI for our clients um so\nvery much just AI yeah yeah yeah yeah\nand so did you start head start with AI\nin mind or was it no we're we're going\nto do coding services and then AI came\naround I didn't I kind of started the\nbusiness more as just I was Consulting\nmyself and what I knew how to do was\ncode and so it was more around like what\ncan I sell in general that I can just\nbootstrap a business and so that was\nreally important to me was not\nnecessarily raising Capital but figuring\nout how to do it on my own the AI piece\ncame in line with it I started using the\nAI to write the code before I started\nimplementing the AI at that time every\ncompany was like I want a chat bot I\nwant a rag implementation I was like I\nneed to learn this and I was very\nmotivated by the money incentivized to\nlearn it if I were going to get paid to\nlearn it and so I basically started\nimplementing it for people knowing that\nthey would pay for it and then building\nthe business up that way for sure so\ntell me about what what are the bread\nand butter projects that yall take on\nright now and like what are your clients\nasking you for yeah so it's really\ninteresting it shifted over time the\nfirst year we were in business it was a\nlot of chat Bots and rag imple\nimplementations we actually still do a\nlot of those but there's a bit of a\ndifferent flavor and then the more\ninteresting projects in my opinion are\nwhen the client has actually tried to\nimplement the AI themselves and they're\nnot seeing as strong results out of it\nor they just want it to do more than\nit's able to do with the implementation\nthat they have and so we'll actually\ncome in and either redesign an\nimplementation or do something new um\nand get very very strong results\nuh with the new implementation so we do\na lot of unstructured data to structured\ndata which is obviously an incredibly\nlarge bucket but document processing a\nlot of web scraping um anything that can\nbe kind of data mapped so one structure\nto another we think in this world\nproprietary data structures are as\nvaluable as the proprietary data itself\nwhen you think about like how you put\nthings together we do a lot of that um\njust honestly the technology is so far\nahead of everyone using the technology\nthat I think it's going to be years and\nyears before people are actually like\ncaught up to the usage of it\nwhich um I get the propriety proprietary\ndata side but what do you mean by\nproprietary data structure and why do\nyou think that is so important so that's\ninteresting I think like there's so much\ndata out there and there's so much data\navailable and the Advent of all the llms\nlike all the data has been now processed\nby the llms and the llm can generate\ndata too as much as you need it and so\nthe structuring is interesting because\nlike as you think about software\nproducts and as you think about value\nwithin it like something being\nstructured either relationally or\nhowever else you want to structure has\ninherent value to it and so if you put\nthe thought process behind what should\nthat structure be and then you can\naccess the data from wherever it is that\nbecomes very very valuable if that makes\nsense sort of I I would love an example\nof those like what do you mean by that\nlike if you think about so every a API\nright every company that possibly has an\nAPI is a certain structure and you think\nabout like lately a lot of software\nthat's built is integration so you're\nintegrating one thing to another and\nthat kind of API wraps the proprietary\ndata structure in the first place if\nyou're then trying to like plug into\nmultiple Integrations there's a mapping\nthat you would maybe do that's like this\nis a common data structure between these\nIntegrations that becomes valuable too\nbecause then you can kind of like pipe\nthings to other things much easier sure\nsure sure yeah and so like as you think\nabout a company that's like leveraging\ndata to make money and whatever sort of\nway and maybe they're taking they're\ninputting the data from all these third\nparty sources doing that mapping into\ntheir structure is almost as important\nas the data sure yeah I mean it's the\nnormalization process to like actually\nmake exactly it's normalization at\nEssence but normalization is so much\neasier with LMS yeah yeah yeah yeah for\nsure okay so you started Head Start um\ntwo years ago what did the team look\nlike now and or what did the team look\nlike then and what's it look like now\nyeah that's a great question so it was\njust me for a year and a half I was able\nhow did you find how' you find clients\nlike how did you get your book um all\ninbound referrals so it was all people\nthat were like I either need the\nsoftware help or AI help part of the\nreason that the book was so broad was\nbecause I had no restrictions on\ntechnology because I could do anything\nwith AI so it wasn't like oh I'm just a\npython engineer or I'm just a JavaScript\nengineer I can only take those projects\nI could take anything because I could\nliterally do anything because I was\ncoding it anyway so that helped make it\nlike very very broad um in terms of what\nprojects I could take on and then I was\nable to scale just me without hiring\nanyone because of the AI because I\nwasn't writing the code the AI is\nwriting all the code so all of a sudden\ninstead of doing like one project you\ncan do five projects at the same time\nand you're like okay I don't really need\nemployees if I have ai um obviously\nwe've shifted away from that and we have\nemployees now but that's the still like\nthe general way that we want to scale\nthe company is in a very very AI native\nway um which kind of goes against the\nway that you would think about like a\ntraditional company of just grow aead\ncount to grow the size totally well\nNicole I don't think you're giving\nyourself as much credit as you should be\nbecause there are bajillions of people\nsitting in the rooms messing around on\ncursor like trying to build stuff but\nyet you've actually built a company and\nyou actually have real clients and so\nwhat is your unique advantage or like\nhow would you assess your skill set on\ntop of just AI coding for you like what\nelse do you bring to the table for that\nyeah that's really interesting so I I\ndon't think that what we do is like what\nwe're doing with the AI is not what's\nproprietary about the business part of\nthe reason I'm happy to like hop on the\nphone and show everyone exactly how\nwe're doing things is I don't think that\nthat is exactly the mode that we have in\nthe business it's incredibly powerful\neveryone should be doing it it's\nincredibly valuable for our clients and\nyou get a huge return on it but I don't\nthink it's like the most important thing\nfor the business ultimately it's like\ncommunication how you get things done\nthe whole thing is hard work and I think\nthat's probably the biggest blocker is\nlike even coding with AI to get projects\ndone it's hard work it's communication\nit's figuring out how the product should\nwork it's getting into the nitty-gritty\ndetails like the AI does a lot but\nultimately so far it only does the\ncoding which can reduce your time\nsignificantly but you still have to do\neverything else well so what what's in\nthat everything else bucket that you\nthink is most important I'd say product\nis a huge part of it right like how\nshould it work and being a thought\npartner on that part of the reason\nclients like to work with us is because\nwe actually help them figure out how the\nproduct should work versus just\nimplementing it exactly how they want it\nlike to work so a lot of clients you\nknow if you just know what you want you\ncan go to any developer and get them to\nbuild it for you but we do more of the\nthought partnership and like well how\nshould it actually work what is the best\nthing to do and like what different\nthings can we build around that and then\nthe architecting of it for scalability\nlike that's the other piece is like sure\nwith AI doing all the coding we can\nfocus on the architecture of the\nsolution and making sure it's the right\nsolution that scales and everything's\nlike just well done and in components\nwith the design system and everything\nthat you possibly need because the AI is\ngoing to do it all that way anyway sure\nI think that's the other benefit is you\nactually get all these other pieces But\nultimately it's kind of communication\nand delivering the end result and a lot\nof that is working with a client to\nfigure out what they actually want yeah\nyeah yeah which is the hard part A lot\nof the time um so you have three people\non the team right now right yes three\npeople a co-founder an engineer and a\nsecond engineer starting week after\nThanksgiving nice congratulations thank\nyou so soon to be four people and so uh\ntwo Engineers or I guess what do their\nskill sets look like so what what skill\nset are you doing versus your co-founder\nversus the two Engineers yeah that's a\nreally good question because we're also\nhiring and we're looking for a very\ndifferent skill set than is like a\ntraditional engineer skill set that you\nwould look for uh between me and my\nco-founder who's also my brother uh I'm\nthe technical one he's the non-technical\nand so we honestly like have very\ncomplimentary skill sets which is\namazing for the business I do all the\ntechnical and a little bit of the sales\nin client coms he also does sales and he\ndoes all the oper AAL so he does all the\ninternal operations he does a lot of\nproduct work too so product scoping is a\nhuge part of our business because we\ncharge flat fee and so we have to get\nall the scope up front in order to do\nthat he does that on the engineering\nfront we look for product Engineers so\nwe look for product Builders we look\ncommunication is such a massive thing I\nthink communication is always important\nas an engineer but lately when we're\nlooking to hire we're actually looking\nfor written communication skills as well\nas verbal communication skills because\nwhen you're using the AI to build you\nactually have to build it use language\nincredibly well to get the output that\nyou're looking for and so we need people\nthat can like look at code and\nunderstand whether or not it's good but\nthey actually don't need the skill set\nto write the good\ncode sure sure yeah it totally does all\nright so four people um how many clients\nright or like what does the what does\nthe portfolio look like yeah so we\ntypically take on like four to five at\nonce um is our kind of typical workload\nbut that's growing as we grow engineers\nand the size and kind of time line of\nthe clients differs dramatically so\nwe'll typically do like a pilot project\nwith a client that's anywhere from two\nweeks to two months and then the client\ntakes the results of that product they\ngo and user test it they use it for\nwhatever they would like to if it's an\ninternal product like their entire\nteam's using it and they figure out what\nmore that they want in that off time\nwe're working on other clients then\nthey'll come back for another project so\nwe kind of have clients coming in and\nout so it's not um contractually\nrecurring Revenue but it's reoccurring\nRevenue because they're Ty having a good\nexperience and they want to then build\nmore um our best client is someone\nthat's like I want to build everything\nand slowly break off chunks and build it\nall for\nthem yeah which is really fun but we're\nfiguring out exactly what is kind of a\nclient per engineer thing look like and\nhow to scale that typically I'm still\ndoing all the client coms which is\nreally interesting and we are keeping\nthe engineers more on product work and\nbuilding out internal tooling to support\nus handling more work at once sure sure\nsure how long is the average client\nengagement that you\nhave average maybe like four to six\nweeks uh they're typically short we um\ndeliver quickly and so we kind of charge\na premium in order to guarantee a fast\ndelivery and so we work with people that\nwant something done yesterday\nessentially and they know exactly what\nthey want and they're ready that's\nawesome that's so nice as quickly as\npossible okay we are a bootstrap\nbusiness and um it's been really really\nincredible to hire out a team of\nus-based like highly paid Engineers just\nyou're all local to New York right yeah\nlocal to New York um it's been really\nincredible to sometimes I don't even\nbelieve it you're like I'm running a\nreal business now\nI'm running a real business from\nhonestly like creating the uh LLC all\nthe way to now we're looking at a bigger\noffice and more employees so that's wild\nmore yeah how many people are you\nlooking to hire um so we are looking to\nhire five more uh for January and then\nfive additional for May um we have a lot\nof plan demand so no joke so you're\ngoing to go from three right now almost\nfour all the way up to 14 in seven\nmonths yeah so really really crazy um\nwe're being very like particular about\nhiring just in the skill set that we're\nlooking for because we're doing such AI\nnative things it's a little bit\ndifferent so we are honestly still\nfiguring out and defining what that\nexact hiring profile looks like as um we\nget into it we're very very interested\nfor May in new grads so people with\ncomputer science backgrounds um who are\nactually not trained in the way of\ncoding so that we can train them in our\nway of coding nice yeah that's very cool\num random question but out of curiosity\nhow do you change your language when\nyou're talking about AI with clients\nversus like technical folks like what is\nthe client tongue you put on to make it\nmake sense to them um I talk very\ntechnical uh sometimes my brother has to\nlike pull me out of it because it gets a\nlittle bit too much in the weeds if I'm\ngetting like really into it but um we we\nare very technical with our clients I\nthink ultimately everything does have an\nexplanation and like people are curious\nabout it and we're we're kind of like\nwilling to explain as much as anyone's\nwilling to listen and hear about uh we\nfind that our most engaged clients\nactually really appreciate it because\nthey even if they're not Engineers\nthemselves are like interested in\nlearning about how the technology can\nwork and how you can use it and in order\nto really understand that understanding\nsome of like the Baseline of it is\nhelpful totally and where do you like to\nplay with regards to um minimum contract\nsize like what what what what starts the\nconversation with you yeah so it's hey\nso after the interview Nicole said hey\nmaybe we don't actually want to share\nthe exact number of their minimum\ncontract value because it changes all\nthe time and no no reason to stake it in\nthe ground but what they did say is that\nthey're charging mid six figures on\nshort-term contracts that's hundreds of\nthousands of dollars which is very cool\nto hear and they're also planning on 10\nXing their revenue in 2025 as they get\nmore Enterprise contracts which I just\nthink is absolutely amazing so let's go\nback to Nicole and hear more about Head\nStart we had to raise it due to demand\nwhen I first started the business um the\nfirst flat fee I did I think was 10K and\nso we just started raising it um\naccording to demand and what's cool\nabout that too is like we get a lot done\nfor that amount like we're getting so\nmuch more done for that amount than I\nthink anyone else would what do you mean\nso what do you mean so much more done\nlike we'll deliver like an entire\napplication build in like um you know\nfrom scratch in four weeks like fully\nfunctional completely ready production\nready MP and so one thing that really\nstood out to me when we were meeting in\nNew York you're like Greg the more\nprojects we do the the quicker each next\nproject gets because we learn from the\nlast one and we have templates and so\nhow does that work with regards to it's\nalmost like your or uh institutional\nknowledge that you're building up that\nmakes the next one quicker so what does\nthat look like yeah so uh we have a Wiki\nin GitHub and we're trying to build that\nout um I'm sure you've heard this with\nllms is like garbage in garbage out and\nso if you're putting like shitty stuff\ninto it you're not going to get a good\noutput but if you put really good input\ninto it you're going to get great output\nand so we're doing similar things over\nand over again I'm usually using using\nthe llm to generate the first version of\nwhat we're doing anyway um so I have it\ngenerate a Wiki alongside that as I go\nthrough its instructions I'm actually\nfixing it so you're almost like kind of\nfine-tuning it yourself when you think\nabout like okay that wasn't the exact\noutput I wanted but this is save it into\nour internal Wiki and then we have it\nnext time that we need to do it to\neither used from our Wiki but better yet\nput it into the LM and say this is how\nwe know to do it and we need to do this\nthing on top of it so it's very Network\neffective of like the more work that we\nget the more we can build out the\ninternal knowledge the more the entire\nteam has access to the internal\nknowledge the more that we can feed that\ninternal knowledge into the LM and get\neven better results out okay wait so I\nthis sounds super interesting I I got to\ndig into this one so it sounds like and\nI want to speak super tactically here so\nthat we can get like to the the Crux of\nit so you have a GitHub Wiki and we're\nliterally talking about a text file that\nsays here are the conventions and the\ntechnologies that we like to use and\nthen the style and Order of them and as\nyou do more projects that gets more and\nmore refined because like as you said\nit's almost like fine-tuning but it's\nnot because it's more prompting more\nthan anything then for the next project\nyou'll copy that Wiki and basically give\nthat as context to the llm right yeah\nyeah so it's in all markdown files\nessentially and then we use cloud\nprojects for everything so basically\nwhatever relevant markdown is relevant\nto the new project you just pop that one\nin alongside the code and then it has\nthat context to go off I want to keep on\ngoing down this but why Claud projects\nover cursor oh I so I love CLA projects\nI love anthropic and Claud in general\nlike I they should pay me they don't I\nthink it's like the greatest thing in\nthe world um I did a case study with\nthem and they have a quote from me in\nthe article saying like anyone that\ndoesn't use it as dumb or something like\nthat like so great um I I don't know\nexactly what cursor is doing on the back\nend and stuff I don't know what their\nsystem prompts are I don't know enough\nabout it to know exactly when they're\nincluding the context into the prompts I\nknow you do the command K from the file\nwe do use cursor I have very Niche use\ncases that I use curser for that like my\nEngineers laugh at me um Cloud projects\nwhat I like about it is it's so\nexplicitly contained so you create a new\nproject sure you write the description\nyou write you put in whatever you want\nin to the content and so I know for\nevery prompt that's going into the\nproject exactly what context is\navailable and I think part of the\nproblem with cursor is like I actually\ndon't know depending on where I'm\nprompting in the tool what context it's\npulling from or if it's the right\ncontext and I was talking about this\nwith Engineers the other day because\npeople always say oh well like if it's a\nreally big code base like how do you fit\nit in and we work in a lot of really big\ncode bases we work with our Enterprise\nclients and and the thing about it is\nlike as an engineer you're not looking\nthrough every file in a really big code\nbase when you're going to change code\nlike you couldn't like you never would\nand so whatever files that you go look\nfor in the file structure to then figure\nout what you're going to code next\nwhether it's a new feature you're going\nto edit something those same files you\ninclude in the cloud project and then LM\ncan do everything so basically you're\ndoing a lot of copying and pasting too I\ndo a lot of copying and pasting\nliterally like my entire work just like\ncopy P copy P copy P yeah yeah yeah yeah\nyeah for sure okay so that's on that\nside um you know I have a sample of\nthose system instructions do can we\ncheck those out yeah so we can I have a\ntest project pulled up that we can kind\nof like look at how we're doing it and\nI'll talk more about a couple other\nthings with CLA projects so yeah so this\nis our projects at Head Start um you can\nsee there's your projects and then\nthere's all projects what I love love\nlove about Cloud projects is that um you\ncan share them amongst your team so we\npay obviously for an Enterprise account\nyou get full Enterprise security baked\nin so you don't really have to worry\nabout like the fact that we do upload\ncode into here um I have this AI\ninterview with Greg project so what I've\ndone here is I have pre-up loaded an\nentire codebase uh that is a test code\nbase that we're working on on and is\nthat the file tree structure in the name\nitself yes so this I do so that the AI\nunderstands where all the fil files are\ncoming from a friend actually used the a\nI to write this script for me it's a\nbash script that I just run in the root\nof my repository that says flatten the\nrepository I can show you I think I had\njust done it in\nhere yeah so it's literally like a\nflatten repost script and run it and\nthen you can\nsee um LS dot dot that I have this\nflattened one and so that just allows us\nlike if you go to Claud you actually\nupload here\num you can see how if it's not flattened\nyou can't like actually upload\neverything to go into here to flatten it\nand then the other nice thing is when\nit's flattened with the file path if you\nthen ask Claude like hey I need to\ncreate a new view can you a new view\nlet's say for settings for the user can\nyou tell me where I should put the new\nfiles and then give me the content of\nthe\nfiles it should look like a\nstandard settings screen I also typo\nlike nonstop with AI it doesn't matter\nat all it's like my favorite so nice so\ndon't have to have\nanything um be perfect and so this is\nkind of essentially what we do so we\nwork with clients that have a ton of\ndesigns we work with clients that have\nno designs and so so if they have no\ndesigns like it'll literally just like\ncome up with everything itself and you\ncan see here it gives you the file path\nso then what I do is I just go back into\nthe\nrepo see in\nchat um where did it want it chat\nsettings page.\nTSX and so we have ID and instead we\nliterally do\nsettings and maybe like if people know\nhow to use cursor to there create your\nfiles I would love to do that I don't\nknow how to do that so I'm always you\ncan do it it's just kind of still the\nwild west because it takes so many\nLiberties and assumptions and just blast\nit all out that you can go deep down the\nrabbit hole it's tough to get yourself\nout of it sometimes really okay so yeah\nwhat I do then is I literally just copy\nthis and I paste it there and I don't\neven really read the code all the time\num I do uh when we're reviewing it for\nclients but when I'm just kind of like\nwriting it like this I think this to\nrename it settings\nnav and then you put that in there and I\ndo like I'll check to make sure\ngenerally it looks right like no\ntypescript errors nothing like that um\nokay so it says create the files I\nalready did that and then it says that\nin the app sidebar I need to add this so\nwe got to go into the app sidebar and\nsee so I just like do this and so you\ndon't have to write any of the code but\nyou do have to figure out let's\nsee nav this one and this here and then\nyou just have to make sure it gets\nimport and then you should be good and\nlet's see if that even\nworked H here it is\nsettings um and then we just have to\nfigure out yeah you gota put the proper\npage in there yeah oh yeah because it's\nunder\nit's just the routing that it actually\ngot wrong cuz it said chat dot settings\nand then we didn't\nactually put it in the Chat Place\ncorrectly so\nsee if we do this it should work and\nthen we just have to hook up proba yeah\nso this is the settings page it came up\nwith look at that that's so wild yeah\nit's crazy and then what's great is when\nyou have figma designs essentially what\nI do is you just dump like here the five\nfigma screens that I need one by one do\nthe screens and then I'll get like first\nscreen over second screen over third\nscreen over and then what I typically do\nis I'll take all of the code that it\ngave me so I'll go through the five\nscreens here and I'll copy paste them\nback and I'll be like here's one screen\nI'm just going to give it these random\nscreens um here's a second\nscreen let's see it's we can be so\nsloppy with our code in llm it's wild\nand by sloppy I just mean like you're\nnot even giving context you're just\ngiving the code yeah I just give it like\nthis and I'll be like between all of\nthese screens is there anything I should\ngeneralize out into components like\nsomething like that uhhuh and the\nscreens I gave were kind of random so I\ndon't even but like if I gave it the\nright\nscreens part of I'm trying to describe\nthis\nright without a good client code example\nwhich we can't share it's hard to\nexplain this but essentially like when\nyou're making new screens you as an\nengineer know generally how the\ncomponent should break out of those\nscreens so like let's just say a\nfriended example of some sort of form\nflow like very standard you have next\nbuttons back buttons maybe you have a\nlittle dot navigation thing you have\nyour various forms you have headers you\nhave descriptions you kind of generally\nknow how those should break out you have\npigma designs that are attached that so\nyou put in all the figment designs you\nget all the code and then you just tell\nthe LM I need this component this\ncomponent this component this component\nand it just pops them out for you and\nartifacts which are so nice because you\njust get the full file copy paste like\npart the reason I don't like cursor is\nwhen like if I highlight something here\nand then I'm editing here it does it in\nline but it doesn't always like get\neverything you need and not everything\nis always like in the right place and\nthen yeah the one thing I'll say on\ncursor too and you can see cursor when I\nscroll over or no I can see your cursor\nright now yeah okay cool um the code\noutput I find in cursor and my engineer\nfinds us too so it's not just me but it\nis just the two of us I've been asking\naround to see if anyone else knows um\nthe code output is better from Claude\nand I don't know if that is due to the\nsystem prompt in Claude which they do\npublish so I don't know if you guys\nyou've looked into that at all but it's\nvery haven't seen it they have ropic\nsystem prompt it's in the API\ndocs my God no it thinks I'm a robot um\nthese are really interesting and so I\nimagine that CS may be using a different\nsystem\nprompt\nthing I'm surprised that cursor wouldn't\nbe able to increase the performance like\nif it's literally just for code yeah and\nCLA I find that the code back from Claud\nis much better I also like very really\nuse chat GPT to code anymore but\noccasionally if I'm running into\nsomething very gnarly I will go into\nchat GPT and get it I find that claw\nartifacts like these like just being\nable to copy paste on the right is much\neasier what I do end up doing a ton is\nand this cursor has this too is like if\nit's a really long file like 500 lines\nor something it'll cut it down and be\nlike you know same code as before here\nsure and I say like hey can you give me\nthe entire file the entire file yeah\nyeah yeah yeah what I also find so the\nlong long files basically like there's\ncertain things with coding that I think\nare going to stand the test of time with\nAI as it gets better and there's certain\nthings that I think will go away I think\nthe IDE as a concept I think I'm like\nquestionable on like I personally do not\nlike working out of the ID IDE in an AI\ndriven world I go back and forth between\nit I'm still in it I get all the\ntypescript stuff and like everything\nlike that that's what I use curser for\nit like fixes all my type errors because\nI'm like just a baby about typescript\nand like don't\nWorks little things like that but the\nIDE is a pattern I'm not confident on\nsmall code files and things broken down\ninto like proper component structures\nand like file structures being correct\nI'm actually bullish on I think that\nwill persist I'm sure you've seen stuff\nwith people where they're like I think\nyou and I had spoken about this once\nwhere it's like does it need to be an\nEnglish language anymore can it just cut\ndown into characters to like shorten\nWindows things like that can imagine\ndifferent ways of coding what I think is\ninteresting is I think English language\nis incredibly important or any language\njust spoken language human language not\ncomputer language um file structure is\nincredibly like structureal\norganizations very very important small\ncomponents are important like building\nblocks puzzle pieces so it's interesting\nvery cool you know it's funny to say\nthat because I've heard an opposite\nopposite opin not an opposite opinion\nbut um a counter opinion which is that\nfile structures are just human\nconventions and that machines don't\nreally care about final constru file uh\nconventions so yeah I don't I don't know\nwhich one it's going to be no I could\nnot agree more that file structures are\njust human conventions llms are human\nconventions they operate on human\nlanguage and human convention and so\nwhat's fascinating about them is the\nbetter you are at human language and not\nmachine language I think the better you\nare eing the\nand that's a very different Paradigm\nthan is traditional in engineering\ntotally which is what I think is very\nvery interesting it's really interesting\nbecause I do think one of the skill sets\nthat I have which has made me able to do\nwhat I'm able to do is like the human\nkind of communication and writing\nability to get what you want out of the\nLM and it's like very very interesting\nif that can evolve away because\nit is evolving more towards it for right\nnow which is fascinating yeah that's\nvery cool um and another awesome thing\ntoo is you had a case study published by\nanthropic on behalf of some of the work\nyou did with one of your clients what\nwhat what what was the story behind that\none yeah so that was really cool um\nwe're just top users of Claude and\nClaude projects uh we love it I think\nit's the greatest thing that has ever\nbeen developed if he took it away from\nme I would really struggle to run my\nbusiness which is probably like the\nhighest NPS you can get for a product um\nand so they wanted to hear about how we\nwere using it and potentially share that\nwith more people and so it was really\ncool because I think we're at least\nusing Claud projects in a pretty\ndifferent way than a lot of their other\ncase studies and so that was kind of\ncool to get um written up for that usage\nyeah they must have a data analyst\nlooking at Claude project usage and\nbeing like dang nocole Head Start she's\nfreaking crushing at least her team is\nyeah no it's um it was a really exciting\nfeature for us for sure yeah that's very\ncool the question I wanted to ask what\nother Pro moves are you doing besides\nthe random of course we're using CL\nprojects but like what other Pro moves\nwith AI en coding are you\ndoing I've been trying to explain it to\nother engineers and figure out what is\ndifferent about how I'm using it one\nengineer told me he's like you trust it\nmuch much more than I would ever trust\nit and so we're I'm taking like sths of\ncode from cloud projects when I'm\nworking on client projects just like\ncopying pasting over and everything's\nbroken down in your typical engineer way\nlike we still have PRS we still review\nthe code it still goes in in chunks It's\nJust Happening that much quicker because\nwe're really relying on a lot of the\ngeneration to write it so I do think the\ntrust level is there um when we we use\nit for everything so like I'm sure like\nwhen we write prompts for the AI\nimplementation for our clients and then\nwe're evaluating those prompts we have\nour own evals product that we built we\nbasically like take the prompt take the\nevals result so basically if it's not\npassing a certain test case we take all\nthe reasoning behind that we feed it\nback into claw to regenerate The Prompt\nso we have like a prompt Loop in terms\nof generating all of that so you have\nnot only your own eval tool which you\nbuilt yourself but you also have your\nown uh like Claud like prompt optimizer\num we do have that and then we also have\nlike an AI uh computer agent that we\nbuil via Cloud computer use so you know\nthe new functionality oh yeah yeah yeah\nthat's cool you already built something\non it yeah yeah um our engineer Tiff she\nbuilt it in like two days and so we've\nbeen using that basically it'll create\nPRS for you um for changes to the code\nit's really good it's kind of like a\njunior engineer um and we've been\nrunning that on all our projects which\nis really really cool so it's been we've\nbeen able to build products in parallel\nwith running the services and then\nleverage those products to be able to\nthen you know the services the computer\nuse are you doing that on a VM or is\nthat literally on her like laptop that\nshe we run it on our own laptop so it's\na script that runs right now um we're\nworking on dockerizing it putting it on\nthe server but there's actually a couple\nnice things about running it on our\nlaptops it like uses the GitHub CLI so\nit'll actually like create a branch and\ndo it from your own terminal which then\nwhat I end up doing with it right now is\nI tell it the change it creates the\nchange it creates the pr and then you're\nalready on the branch in your project\nlike in cursor and so from cursor then I\ntest it myself like locally and so I can\nstill do the testing we don't have the\nAI agent doing the testing yet because\nthat's obviously like a little bit more\ncomplex yeah and then I can make the\nchange either personally or I'll have\nthe AI agent do it and so we have that\nkind of tight feedback loop within it\nand what's so cool is like we're dog\nfooding our own product yeah so we can\nimmediately like just you know this\nthing didn't work like it created an\nempty file and it got caught in a loop\nor whatever else but computer use is\nlike very powerful yeah that's wild um I\nwant to ask about tool stack it's always\ncool to hear which tools people have in\nin case there's one that I'm like isn't\npart of my routine yet so what's in your\nAI development stack and what are you\nusing besides Cloud projects and besides\ncursor Cloud projects cursor uh chat gbt\nGitHub obviously huge fans of GitHub um\nwe use we build a lot of like different\nproducts for people so we use react\nnative for mobile apps\num we use any kind of full stack\napplication we use versell a lot for new\nprojects um I like using versel to spend\nthings up um a lot of node on the back\nend or python on the back end is\ntypically what we use we work within any\nclient's existing codebase so we'll use\nwhatever they're using across the board\nbut when we're doing n new typically\nit's like reactor react native front\nends typescript nodejs python backends\ndepending on what we're building\nmicroservices we're typically using\npython 4 um different databases stuff\nlike that\num yeah that's kind of we we keep light\nlike we just paying for claw cursor open\nAi and then like the usual business\nstuff slack yeah\nEtc what um that doesn't sound like a\nlot what are your what are your margins\nfor your business oh we just I mean we\npay engineer salaries plus like $60 a\nmonth per employee for like all the AI\nlike nothing basically all our clients\npay for the um for their AI costs and\neverything so yeah and we have a little\noffice in New York so those are\nbasically the costs for the business um\nI still think it's insane that through\nopen Ai and inop you can get this tech\nfor 20 bucks a month or 30 bucks a month\nwhatever it is absolutely wild and that\nyeah that's crazy because like you're\nnot even paying the API costs for your\nCloud projects right you're justay the\nbusiness Enterprise outside we pay the\nAPI cost for our own test projects so\nthat's the but honestly it's pretty much\ndriven to zero like we do a lot of\nquoting for our clients on how much the\nAI will cost them and it's so like it's\nlike point\n0000000000 two then multiplied out like\nit's hard to get numbers up there yeah\nyeah yeah for sure for sure um so one\nthing that's come across in this\ninterview is you have a lot of technical\nconfidence which is very cool it's like\nany project just brings it on and we'll\ngo do it um what types of technical\nprojects were you steer away from like\nwhich one are you saying no to that\nclients bring to\nyou um like below the minimum fee\nbasically uh we're we're not afraid of\nanything we try to take on the hardest\nprojects that we can we believe and it's\nwritten down on like our strategy board\nthat access to hard problems is like a\nproprietary business value for us um so\nwe think that having that access is\nincredibly powerful and we definitely\nDon't Stray away from it sometimes it is\nscary because we have to do really hard\nprojects that we don't necessarily know\nhow to do in the onset but we are able\nto figure it out and have that\nconfidence but yeah typically like we\nwill do anything that people think is\nimpossible and we like to do that um one\nCU we can charge for it which you know\nas we should if people think it's\nimpossible but also because we learn\nfrom it and we think it makes our\nbusiness more powerful that's wild and\nthe clients are happy because we're\ndoing essentially be impossible for them\nyeah what else is on that whiteboard of\nvalues that your company has oh yeah so\nwe're very values driven uh company\nvalues are Simplicity patience and\ncompassion I pulled them from the Dow\nday Jen the Steven Mitchell translation\num it's one of my favorite books and so\nthat's kind of like the core values but\nin terms of like access to like propri\nhard problems being proprietary like\nthat's where we think about like our\nMoes like what is more of like the power\nfor the business and it's not how I use\nCLA even though how I use CLA is like\ndriving the business forward it's very\nmuch kind of like you know doing good\nwork for our clients is number one um\nclient NPS is the thing that we care\nabout the most our clients understanding\nthe type of work that we can do um and\nthe quality of work we can deliver for\nthem is number one uh most importantly\nefficiency of the business so we track\nrevenue for per employee that's really\nimportant as we grow the business uh it\nsounds like that is just insane right\nnow it's high right now yeah and we\nfigure out how to continue to scale but\nbasically because we're a Services\nbusiness and we're also investing in a\nproduct that doesn't scale like every\nother business it's more jumpy because\nas we take time to build product we'd\nhave less time for services so we're\nfiguring out how to like even that out a\nlittle bit as we grow um but yeah we\nthink communication client NPS um access\nto hard problems being able to solve\nhard problems um being able to do things\nin a repeatable way creating Network\neffects within the business and like uh\ngood data like good data in terms of\nlike how we use the\nllms good code is kind of proprietary\nand so yeah we think about it a little\nbit differently than like how we use\nllms as prior proprietary sure um what\nabout businesses that are two to three\nyears ahead of where you are right now\nor that you want to be or two to three\nsteps ahead what do those businesses\nlook like I mean I think this is just\ngoing to make me sound like a crazy fan\ngirl but I think enic the greatest\nbusiness out there right now um I think\nwhat they're doing is incredible and\nlike the product itself that clot is is\nreally really powerful no I mean like\nwhere do where do you want to go though\nlike where do you want to take your\nbusiness like in two to three years oh\nthat's a great question um we're trying\nto grow in an AI native way so we want\nto continue to hire but not hire uh to\nthe scale of the client work we want to\nhire and then train up a team that can\nthen do that work exponentially we think\nthat the progress of AI is inevitable\nand we're building into that uh\ninevitability as a company and so\nwhether or not the AI will be able to do\nthis work completely next week or\nwhether that will happen in two years to\nme it's inevitable even if it's 10 years\nand so whatever that timeline is we're\nkind of just making sure that we are the\nbest at using the AI and we're best at\nimplementing the AI in the business and\nif we do those things I think the\nbusiness will continue to grow yeah well\nwhat about products or SAS is that in\nyour future so the way we're thinking\nabout products right now as we have them\ninternal products that we've built um we\nare going to continue to build those and\nwe want to be the users of our products\nfirst and foremost we're not trying to\nbuild products for other people right\nnow um we're trying to build products\nthat make ours us more efficient across\nthe board if we can do that and we've\nproductized that in a really powerful\nway and we feel like maybe selling it\nit's an option but it's not priority\ncool beautiful Nicole that was fabulous\nthank you very much for joining us today\nthank you Greg it's always great\nchatting with you",
  "transcript_chars": 43813,
  "ingested_at": "2026-05-15T04:40:25.100749+00:00",
  "source": "channel",
  "yt_meta": {
    "view_count": 32167,
    "like_count": 976,
    "channel_id": "UCyR2Ct3pDOeZSRyZH5hPO-Q",
    "categories": [
      "Science & Technology"
    ],
    "tags": []
  }
}