{
  "video_id": "-B0zaWoPEFs",
  "channel_slug": "deeplearningai",
  "channel_handle": "DeepLearningAI",
  "title": "AI Dev 26 x SF | Panel Discussion: Future of Software Engineering",
  "duration_seconds": 3309,
  "url": "https://www.youtube.com/watch?v=-B0zaWoPEFs",
  "upload_date": "20260520",
  "transcript": "Hello everyone. While we're getting\nsettled, I'm just very happy to announce\nthat today the topic of our conversation\nis going to be the future of software\ndevelopment. Uh my name is Marina\nMcGillo. I run a podcast called Silicon\nValley Girl where I interview [laughter]\npeople who are building the future of\nAI. And I've been asking this question,\nwhat jobs are disappearing right now?\nAnd when I talked to Reed Hoffman, who's\na legendary investor, he said developers\nare still going to exist, but instead of\nwriting code, they're going to be\nmanaging multiple agents. Then I talked\nto the founder of Replet Amjad Masad and\nhe told me that you know idea and\nmarketing are key product building is\ngetting more automated. So we get so\nmany different opinions and we live in\nthis exciting time of change. So today\nwe have an amazing panel and an amazing\nconversation. I want you guys to\nintroduce yourself first and then give\nme a number from 0 to 10. How bright is\nthe future of software development? Zero\nbeing very dark, 10 being very bright.\nUh, hi, I'm Joe Ree. I am a data\nengineer, data architect. Uh, I wrote a\nbook called Fundamentals of Data\nEngineering and also did the uh deep\nlearning uh data engineering\nspecialization with Andrew. Um, and I\nwould rate the um the brightness scale\nprobably an eight, an 8.5. We're allowed\nto use a decimal.\nThat's good. Okay,\nperfect. So Dan Maloney, CEO of Landon\nAI. At Landon AI, we're focused on\nmaking the world's documents computable.\nSo we uh we spend a lot of time every\nsingle day thinking about how to take\nthat messy world of documents and bring\nthat into structured data so Joe and the\nteam can handle it from their side. Um I\nI'm actually quite optimistic. Maybe\nit's because my uh co-founder partner in\ncrime Andrew uh we believe a lot in\nbuilders and developers. I see the\nnumber I I'll put it up in that 8 to\nnine range. I think the skills around\ncritical thinking and other aspects are\nstill essential. The the role will\nevolve. We'll go into it more here, but\nI'm I'm still quite positive. Again,\nit's a yes, but uh type dynamic.\nYeah. Uh Richmond, director um of AI\ndeveloper experience of our Oracle AI\ndatabase and over Oracle AI database, we\nare focused on being that unified memory\ncore for AI agents that you guys are all\nbuilding. And in terms of my optimism on\nthe future of software engineering, it's\nreally based on who is exactly who's\ngoing to be doing the actual engineering\nuh two to five years from now. So my\noptimism will be on the spectrum maybe\nseven out of 10.\nHi everyone, Mika Katasta, president and\nhead of AI at Replet. Replet is the best\nweb coding platform for knowledge\nworkers. So, as a matter of fact, I will\nhave to give a 10 out of 10 for how\nbright I think is the future of software\ncreators, given that our company goal is\nto be in the hands of a billion people.\nSo, we're going to see an amazing amount\nof software creators appearing in the\nworld in the next few years and way more\nsoftware being created on a on a daily\nbasis.\nSo, what's your number?\n10 out of 10, of course.\nOh, nice.\nDoesn't get brighter than that.\nOkay, but let's start with you. So, as\nwe see software engineering is changing.\nCan you name the main things that are\ngoing to change in the next three years\nfrom what you're seeing at Replet?\nLet's divide it into I think the 50 plus\nmillion software developers in the world\nhave received basically superpowers in\nthe last few months and I'm sure I'm\npreaching to the choir whoever has ever\nwritten code in this audience. They're\nusing coding agents today. I'm sure they\nare moving much faster and being more\nproductive. That is not going to stop\nhere. I think we we just started and the\nentire software development life cycle\nwill be revolutionized by coding agents.\nWhat personally I'm even more excited\nabout is the fact that the remaining one\nor two billion knowledge workers in the\nworld who today were waiting for their\nengineering team to have bandwidth to\nbuild what they wanted now they can get\nit done on their own. That's the real\nrevolution that is happening that is\nprobably a bit less flashy than the fact\nthat developers are moving faster\nbecause that's what we mostly focus\nabout in the industry. But there's going\nto be an avalanche happening in the next\nfew years because you know a product\nlike Revit.\nSo how does a beginner developer become\nsenior if a lot of beginner work is\ngetting automated?\nThat's a great question. I think every\nnew grid ask me exactly the same when\nthey when they join the company. The the\ntruth is when you use a coding agent on\na daily basis, the amount of complexity\nyou're exposed to is much higher than\nbefore. So when you when you join a\ncompany back in the days as a junior\nengineer, usually we tailored a small on\nboarding task for you and then you you\nonly saw a minimal amount of complexity\nover time. Um as a new junior software\nengineer, you're put in front of the\nentire complexity of the system and\nyou're going to be burned very quickly.\nlike you're going to be making mistakes\nin the first few months. It goes without\nsaying, but that is the kind of\nexperience that quickly elevates you\ntowards becoming a senior. It's just it\ncompresses the timeline that brings you\nfrom being a junior to someone who is\nexperienced enough. It's going to change\nthe way in which we have to learn. It's\nit's a totally different relationship.\nYou're not going to have any more only\nhuman mentors. Actually, the AI is going\nto mentor you to elevate yourself. But\nif you know how to move at a fast pace\nand you have curiosity, your career\ntrajectory is gonna be speed run\ncompared to the past.\nSo basically AI is raising the bar for\nfor everyone 100%.\nOkay, Dan, my next question is for you.\nGoogle says 75% of their code is now\nwritten by AI. In five years, do you\nthink they're going to need less\ndevelopers because it's more and more AI\nbased? So I I think you know I can even\nlook internally at landing AI and how\nmuch over the past 18 months we've gone\nfrom uh very little being developed by\nyou know agents to now we just did a\nsurvey internally and it was up near\n100% as far as that the development as\nfar as internal um I think that will\ncontinue to grow but what's really\ninteresting is as we've now done a great\njob at you know more of our code being\ndeveloped by whether it's quad code or\nwhatever you're using or replet or these\nother aspects I I think you kind of hit\nother walls as you go down downstream.\nSo when you're starting to look at\ntesting, when you're starting to look at\ndeployment, other aspects and your\ndevelopers are able to do in a sense\neven more where you know they might have\nuh various agents or supervisors working\non top and so they can actually do more\nuh an individual developer can actually\ndo much more. So I think it it will\ncontinue to grow as far as the amount of\ncode that's written by uh you know\nindividual developers and I think\ncompanies as a whole will head there\nmore. But again, I see it as an\nopportunity for developers to not have\nto spend as much time and actually code\ndevelopment and, you know, PRs and and\nreviewing of code and so on and they can\ngo on to, you know, more of the outcomes\nand the downstream um and really even\nmaybe starting to blend a little bit of\nthe engineer and product management uh\nin a sense kind of together.\nGot it. And we all see a lot of posts on\nX where somebody says, \"Oh, my agent\njust built this company and launched it\nwhatever.\" Do you think in general we're\noverestimating what agents can do?\nBecause from my own perspective, when I\nwork with an agent, I still need to work\non the input and maybe adjust it and\nthen the output is not 100% accurate.\nWhat would you say?\nYeah, I I think so. First of all, people\nare still very much in the\nexperimentation mode. Even some of the\nbest developers I know at at great\ncompanies here um that are, you know,\ntoken maxing to the extreme and have\nagents running all night for them. I\nmean, as they go through it, they they\nlearn what it can do, what it can't do,\nwhere to set up guard rails, etc. Um,\nsome hope to actually not have to work\nnecessarily as hard, uh, because the\nagents are going to be doing everything,\nthen they find out they're actually even\ndoing more and they have to be more\nengaged. So, I I think right now it's at\na very interesting stage where, yeah,\nyou will, you know, um, work with your\nagents, have them do more, but there's\ngoing to be a lot of experimentation\nback and forth to kind of optimize that\ndynamic. But I I'm I'm very excited\nabout what it can do and you know as as\nnot just individual agents but agents\nworking with other agents and aspects\nand then how that human will be\ninvolved. So I think over the next\nseveral years that's going to be um a\nreally exciting dynamic where a lot of\nuh you know innovation and other aspects\nbut right now the one thing I'll say to\nwhether you're a junior developer,\nsenior developer, I think just getting\ninvolved in it, jumping in starting to\nuse the latest tools is the key piece.\nAnd I I'll almost say this story that I\njust heard from a I'll say a junior\ndeveloper who I think is pretty seasoned\neven though he's a young developer. he\nwas saying, you know, at his company,\num, some of the more seasoned people\ncoming in, whether they were in top\nacademia, whether they were from great\ncompanies like Google and and others, he\nactually knew a lot more than they did\nhere early on. And he was actually\neducating them, not on how to be, you\nknow, their 30 years or 20 years of\ngreat software experience, but how to\nget them up to speed on how how quickly\nhe as a younger developer had adopted\nthe AI native tools. So, I think there's\nstill a blend for you, almost going back\nto your question, there's still a blend\nfor both as we uh as we go forward.\nAny horror stories of agents breaking\nthings in your company?\nI'll hold on that for right now. I'll\nhold on that for right now.\nOkay. Because we we keep hearing more\nand more of those. So, people that's\nback to the notion of overestimating\nwhat agents can do and giving them too\nmuch to work with.\nYeah. I I'll just say this at a high\nlevel. Um I horror story might be the\nthe wrong word, but uh I I will say\nthis. I I think we're stressing a lot of\nCFOs out right now and heads of finance\nas the developers are uh playing with\nwhatever you know tools and agents they\nthey might need to uh to do to get their\njob done. So I think there there's kind\nof a reckoning going. I I even saw\nsomething with uh you know an article\nabout written with Uber and and kind of\ntheir their work and their pieces and\nthat the kind of blowing through budgets\na year budgets in in a quarter. So I\nthink uh that'll be an interesting\ndynamic how that works out over the next\nyear.\nAbsolutely. Uh Rich, my ne my next\nquestion is for you. So when we talk\nabout agents, they obviously need a lot\nof memory to work with. What does that\ntechnically mean?\nWhen we talk about the memory for AI\nagents or computational systems that\nyou're all building, the easiest way to\ndescribe it is if you think about humans\nand our ability to recall information\nand store information using our brain.\nNeuroscience is still a active area of\nresearch but from what we know from\nneuroscience that there are different\ntypes of memory in the human brain and\nthe way we store information differs\ndepending on the type of information\nwe're storing. The same logic actually\napplies to agents as well. We can\nactually start to model memory and\nstructure memory in a way that we expect\nthem to be retrieved and stored and\nevolve over time. So the common very\nsimple um types of memory are procedural\nmemory, working memory, um episodic\nmemory and semantic memory. Those are\nvery easy to understand and uh with\nshameless plug we have a course on deep\nlearning AI on building memory aware\nagents that teaches you all of this. But\ngenerally we are trying to\nbuild systems that can run for long\nhorizon task and actually can long for\nhours and that could run for several\nhours in days. So when you actually when\nyou're building that system you need an\nefficient way of recalling information\nretrieving information and more\nimportantly forgetting information and\nthat's what agent memory is. uh any bad\nexperiences from your point of view when\nmemory wasn't organized in a good way\nand the agent messed things up? Do you\nknow what? This one is a bit personal,\nbut I changed I joined Oracle about\nmaybe six months ago and interacted with\nsome of the tools. I'm not going to name\nthe tool, but interacted with some of\nthe tools that we all use on our daily\nbasis. After 3 months into a new role\nand several conversation, it's still\nfull. I worked at my old at my old job.\nSo, one thing I expected was from the\nconversations I'm having with this\nintelligent entity, I would expect it to\nupdate his uh data and update his memory\nabout um about me, but it didn't. So, it\njust shows that memory is not solved.\nIt's not a horror story, but it shows\nthat there is still a lot of work to be\ndone in the space of agent memory and\neveryone can play in this space from a\nsingle solo developer to all the\nhyperscalers of Frontier Labs. Yeah,\nit's just fascinating how many\nopportunities are opening up because a\nlot of people are like, \"Oh, AI is bad.\"\nBut then we're talking here, there are\nso many problems that are just arising\nthat need to be fixed and that are great\nopportunities. Um Joe, my next question\nis for you. Let's talk about data. So\nall of this depends on data, but data is\nvery messy. What are companies getting\nwrong about data in this new AI era? I\nthink the things that uh companies are\ngetting wrong is feeling like they can\nspeedrun into the future and ignore\ndata. Um I just he was talking backstage\nwith somebody about this and it there's\na certain sense with AI and agents now\nthat uh well we can just put agents on\ntop of our database and just talk to it\nand that'll be that. Now of course when\nyou actually try this for real it\ndoesn't really work as cleanly as you\nfind. And so what what's interesting is\nuh when I talk to practitioners and in\nmy own work uh there's a sense that we\nneed to go back to the fundamentals of\nuh clean data for example so data\nmodeling quality governance all these\nare getting a second look uh because of\nAI and I think one of the themes right\nnow is it's helping us move fast but\nspeed also means you have the feedback\nloop to uncover problems that maybe you\ndidn't before right and so I think that\nwhat's going on now is it's uh shining a\nlight on probably a lot of the sins of\nthe past that we've sort of swept under\nthe uh the rug or stuffed into the the\nproverbial closet so to speak and for\ndecades have ignored um now you don't\nget that luxury because AI because it\nmoves fast it exposes a lot of the weak\nlinks um in your infrastructure and your\ndata and um I mean who's seeing this\nright now in your own companies when you\nYes. Uh exactly. So, so I think we're\njust at the beginning of this and so but\nI think that the where people would get\nit wrong is thinking well um we'll just\nslap Claude on top of our um production\ndatabase and uh ask questions and have\nit uh you know use that as a system of\nrecord. You certainly can. I'm not going\nto stop you. Go for it. Um token max to\nyour heart's content on that. Um but and\nit was interesting. So I just said I\njust finished a survey uh with my my\ncommunity on uh data and um and so forth\nand one of the interesting things was uh\none of the anecdotes was a chief\ninformation officer had uh didn't want\nto work with their data team to build a\ndata model and a semantic model uh to\nmake the uh queries clear. So he just\nput Claude code on or claude on top of\nthe uh data warehouse and just uh bypass\nhis entire data team like well I want to\nmove fast and build dashboards off this.\nBye-bye. And uh yeah, it turned out to\nbe an absolute disaster. Um I don't know\nif this person will have a job. Uh but\nthat's a story for another day. Uh\noh. So when it comes to AI systems in\ngeneral, what do you think? And that's\nthe question to all of you. What is the\nmain bottleneck right now? Is it memory?\nIs it data? Is it the structure of old\ncode?\nIn my experience, it's kind of a lot of\num like all the above. But if you dig\nunderneath that, right, they if you look\nat the kind of the root causes of a lot\nof the um uh code, for example, or bad\ndata or all this stuff, it stems back to\nthe sense that um there's an old saying\nthat we we we don't have enough time to\ndo it right, but we have enough time to\ndo it over,\nright? And I think AI is sort of\naccelerating this sense where we can\nmove extremely fast at things. But um at\nthe same time when I look at my uh my\ncommunity and and when I talk with uh\npractitioners and and leaders, there's a\nsense that we've always had time\npressures to do everything. And now AI\nis actually making it worse in some\nsense, right? So now because there's a\nsense that we can go a 100x, a\nthousandx, a millionx faster than we\nused to. Therefore, let's do that.\nRight? Right. And I can you can imagine\nwhat what what what's happening. I'm\nsure you see this in your own works. I\nthink that's that's the biggest thing is\nit's it's interesting. We're at this\nweird paradoxical stage where we could\nuse AI for um the you know to to help\nmitigate the things in the past. And\nit's it's an interesting tension where\nnow leaders are like well we have a\ntoken budget that will blow through in\nthree months instead of a year and let's\num let's use that and and produce more\nstuff. So it's it's a fascinating time\nright now. Dan, what do you think the\ncompany should be doing to prevent\nthings like this from happening?\nYeah. So, um, you know, when I think\nabout, you know, where where we from\nlanding AI's point of view are kind of\nseeing challenges in the bigger picture,\num, you know, I I still think there's a\ncouple dynamics where I'll say as far as\nwhen we're writing things like code,\nI'll say we destroy things very often\nand kind of can rebuild them from the\nground up really quick. I think that you\nknow Andrew's Andrew's has kind of got a\na great kind of saying around\ndatacentric AI and how critical data is\nand and quality in you know quality out\nand that dynamic. So I think for us you\nknow especially in the world of like\ndocument processing and these other\npieces it it's still a little bit of a\nlong pole in the tent having the you\nknow the right data spending the time\nwith the a lot of the the data labeling\nall the other pieces in and this\nprocess. So for us, I think that's kind\nof a key area where we end up saying um\nyou know, from from where we're focused\nin our pieces, that's where we we spend\na lot of our time. But, you know, I I\nthink as we as we um you know, evolve in\nthis dynamic, I'm sure other roadblocks\nand other pieces will will set up uh and\nand kind of raise their their uh ugly\nhead. But for now, I think that's kind\nof where we say, you know, that's a\nthat's an area where we have to put a\nlot of emphasis and do it right the\nfirst time or the results downstream end\nup being uh quite poor when and and\naccuracy means so much for what we do.\nWhen you need that high accuracy, you\nneed to do a great job up front or\neverything downstream ends up having a\nnegative impact.\nGot it. Richmond, what do you think\nin terms of your original question as to\nthe AI systems we're building today?\nWhat are the bottlenecks? And\nis in a very interesting position\nbecause it's it works with pretty much\ndifferent companies of different size\nand a large number of uh companies as\nwell. So we see a different dimension to\nthis problem of complexity in building\nAI systems. So the first one is\nthe interface. The interface of building\nAI system is changing so rapidly. It's\nnot settled. um people are either doing\nall code they're more pro code or they\nwant to use low code or no code\ninterfaces we see that changing people\nwant to use MCP people want to use CLI\nand it's all about I think the\nbottleneck is the interface in the sense\nthat because it's not settled the\nanother bottleneck is actually role\ndefinitions um the again the the role AI\nengineer didn't exist maybe 3 years ago\nand we see a lot of people trying to\nunderstand where they fit in this new\nagent world in terms of the jobs that\nthey have to do, how they have to\nupskill. So we find ourselves actually\neducating a lot of developers um a lot\nof business analyst and people within\ndifferent business units, people like\nyourselves on what the future looks like\nand what they could be doing in the\nfuture. And um one of the one of the\nroles is uh I see emerging is AI memory\nengineers. And these are people that are\nreally focused on making sure the agents\nremember the right things and remember\nefficiently. Another role I was\ndiscussing with a few people in the\naudience today was um agent architects\nwith the emergence of um local tools and\nvisual gueies to build agents. I see a\nlot of developers and IT architects\ntransforming into agent architects where\nthey actually are abstracted away from\nthe code but can build the building\nblocks and create all of this workflows\num that automate um previously manual\nprocesses. So just to recap the\nbottlenecks that in AI systems we're\nseeing today are the interface right how\nwe interact with the tools but also the\nrole definition how we understand the\njob to be done and how we bring people\nalong and upskill them but the\ntechnology is moving at such a fast pace\nand that is not a bottleneck the the the\nmodels are increasing in terms of the\nreasoning capabilities the tool set is\nexpanding so the the possibilities is\nalways exciting\nYeah. Uh, we we should probably the next\nquestion would be and I'll give you some\ntime to think about this. I love how you\nmentioned the job of the future. When I\ntalked to the CEO of LinkedIn, he told\nme last year they've had 1.2 million new\njobs that are AI related that just\nappeared on the platform. So my next\nquestion would be about maybe you have\nan idea what's the job of the future is\nMichelle and uh getting back to the\nprevious question what do you think\nwhich layer is the most important is\nbecoming the most important when it\ncomes to AI\nI'm going to connect to what you said\nabout education before the I think the\nthe trickiest aspect of AI system\nespecially when it comes to coding\nagents is understanding and conveying to\nthe user what's the ceiling of that\ntechnology and the reason is the floor\nat this point is really clear to\neveryone. The access to creating\nsoftware has become basically the\ncreative using a chatbot. You write in\nnatural language what you want to create\nand things appear magically in front of\nyou. The ceiling is much more tricky\nbecause that's where you keep pushing\nthe boundaries of what the system can do\nand that's when you start to be really\nfrustrated because you know the AI\ndoesn't respond in the way you want.\nMaybe the memory layers are working as\nwell as it should. Matter of fact,\nsometimes you fix a gnarly bug with an\nagent and you try again a few days later\nand it's not going to do a good job once\nagain because there was a lot of sweat\nput by the user in making that happen.\nSo I think that if we had a better way\nto convey in product how far you're\nsupposed to push an agent and when to\ngive up it will it will lead to just a\nbetter user experience and I would say\neven level of adoption in the industry.\nUm I consider that the trick as one\nbecause I don't think it can be fixed\nper se product. I think it's a mix of\neducation. And I think it's a mix of\nmaking models better and as a as the the\nentire industry, we have to become\nstrictly better at that over time.\nOtherwise, I'm afraid there's going to\nbe like a big push back from our users.\nI absolutely love this approach of\ntrying to push the ceiling and\nunderstand where the ceiling is,\nespecially when it comes to AI because\nsometimes we just don't realize how\ncapable the system is or maybe how\nuncapable it is.\nUncapable. Exactly. Okay, let's talk\nabout a jobs of the future because\ninside your companies and what you're\ndoing, you're seeing so many tasks that\nyou're trying to give to people who are\non your team. So, what are those tasks?\nWhat should people be learning? And what\ndo you think is the job of the future?\nI don't know. I don't know that I could\nreally point to a job maybe a job title\nof the future per se, but I could set a\ntest.\nMaybe the characteristics of somebody\nwho's going to do really well in the\nfuture. I think maybe is how I would\nlook at it. And I think somebody who's\ninnately curious about the world and\ncontinuously learns um and I think\nironically takes the time to slow down\nand think about problems and how to\nsolve them, I think is going to do\nreally well in this future because you\nhave a basically an infinite workforce\nright now with agents. But are you going\nto be like the Michael Scott version of\nthe manager of those agents or are you\ngoing to um maybe be a bit more\nthoughtful uh about uh what what you\nwant these agents to do? So I think\nthat's one of the things I've been\nthinking about really is really like\nwhat is the the archetype of a person\nwho's going to succeed when you have\nbasically infinite number of employees\ndoing the work. Um but those are some of\nthe characteristics I would say is just\nget really good at the domain that\nyou're trying to solve in become\nbestin-class at that. Um obviously learn\nthe technology. Uh but you know I would\nI would actually say one thing I've been\nworking on right now is actually just\nlike doing a token minimization with my\nlife. Like I traveled here from Salt\nLake City with my phone and a remarkable\ntablet and a hardcover book. Why would I\ndo that? Because I actually want time\njust to sit and process cycles at human\nspeed versus um trying to just\ncontinuously token max until I pass out\non my computer at 3 in the morning. So I\nthink that's um kind of where I feel um\nyou know is sort of the tension right\nnow. So\nthank you Dan. What do you think?\nYeah, you know I I almost would say a\nsimilar dime. It's hard to exactly guess\nthe um the exact role or or or titles of\nthe future in my mind, but I think\nthere's a few things that are that are\nclear and and I've even seen it whether\nwe're talking with friends at other tech\ncompanies or even in our company. Uh\ncritical thinking is is going to\ncontinue to be essential in this\ndynamic. Um you you have new tools, new\ndynamics, new ways to solve problems. Um\nbut it's not always crystal clear about\nalways how to do that. So you'll still\nneed that that deep understanding. I\nthink just naturally developers and\nbuilders have um the curiosity uh the\nability to deal with change very quickly\num they are solution solvers etc. I I\nthink that will be an essential piece of\nthe pie and then you know the the things\nthat will be automated. I think of less\nautomating exactly what a human does and\nthe role that they're doing. I I think\nof tasks being able to be completed. So\nthere are things that every role um even\nas a CEO that I would love and by the\nway we've taken advantage of building a\nlot of automation but you know there\nthere's tasks that I want to be\nautomated in what I do but we haven't I\ndon't think the AI has gotten good\nenough to actually complete and do that\nwhole role on a on a individual day in\nan individual dynamic. So I I like to\nthink of you know kind of the roles of\nthe future where you know as we've you\nknow the way that you know houses were\nbuilt uh you know hundreds of years ago\nto way they are now and with CAD and the\ndifferent tool I think these are all new\ntools whether you're talking about\nreplets or you know what Richmond and I\nwere talking about some new capabilities\nfrom Oracle etc. you'll have to learn to\nuse this new tech. It's moving fast.\nThere's never a bad time to jump in.\nEven if you feel behind, you can\nactually the tools allow you to get up\nquite fast. Um, you know, I'm even\nalways talking to, yes, my co-workers,\nthe engineers, uh, even my son, etc.\nthat are all, you know, trying out new\ntools. So, a lot of that sharing back\nand forth. So, to me, um, as long as you\nhave that those critical skills, that\nwill kind of, uh, allow you to, you\nknow, go into those jobs of the future.\nThat's kind of how I position that\ndynamic. I like that you mentioned this\nfeeling of falling behind because again\nwe're surrounded by social media and all\nthe posts where AI agents are just doing\neverything for people when in reality\nthat might not be true. So it's always\ngood just talk to people around you ask\nwhat they're doing. I was just in New\nYork and I was talking to the CEO of\nDolingo. He's like you know we did a lot\nof AI code and then we spent so much\ntime debugging it so we actually\nstopped.\nYeah.\nBecause of that. So it's also once you\nget out of our Silicon Valley bubble,\nyou do realize that people are not that\nadvanced yet. So if you ever experience\nthis FOMO, start talking to people\naround you.\nYeah. And I won't go deep into this now,\nbut I think that's a whole another topic\nyou just made where sometimes companies\nare already have products that they\nmight have developed, then they're\nbringing in some type of agents or, you\nknow, whether it's cloud code, whatever\nyou're using, uh, and they they hit\nwalls in this dynamic. I think there is\na new approach about where to start, how\nto do things, how easy it is to get rid\nof technical debt, destroy things, start\nback over. Whole another topic, but\nyeah, you you said some interesting\nstuff there.\nYeah. Yeah. Richmond, what do you think\nin terms of the\njob of the future? So, one thing I tend\nto or I don't like doing is making\npredictions far out in terms of AI. So,\nI'll talk about the near future. We\ncould look at it within maybe a six\nmonth to one year time frame. I'm very\nhyperfocused on AI developers and if you\nactually think about the term software\nengineering it came about\nin the 60s right in a in a NATO\nconference when we had a bunch of people\nthat wrote programs they came together\nto discuss the complexity of building\nthis program and creating a discipline\ncalled software engineering to make sure\nthat people take it very seriously and\nobviously we all studied it in\nuniversities we study on a day-to-day we\nbest practices and principles and\nmythologies of software engineering and\nall of that was so that we can build\nreally good softwares and computational\nsystem and abstract away the complexity.\nNow today we had we have another layer\nof abstraction. So in terms of uh no\none's writing machine codes because we\nall learned programming languages like\nPython, Java, C and that abstracted the\nmachine the assembly language. But today\nwe have another level of abstraction on\ntop of the language which is the coding\nagents which is they allow us to\nactually build the systems without\nalmost some people not interacting with\num the programming languages\nparticularly. So vibe coding is a thing,\nbut honestly, you're not going to vibe\ncode your way to production. Um, if you\ndo, you're going to you're going to find\nout that it's very difficult. But what\nI'm trying to get to is in the future,\nwe need to start focusing on where is\nthe where where is the trend of the\nabstraction and that would define the\njob that we need to do. I see a lot of\nAI developers today writing less code\nand going into more management of this\nagentic systems but and this is very\nimportant\nbut needing to understand the code that\nis written right you also need to\nunderstand the code that is written\nbecause maybe 90% of the time it will\nwork but the 10% of the time that it\ndoesn't you need to deep dive and start\nto go through the code and understand\nwhat changes you need to make or what\ndirections you need to give to change um\nto change some code in some to make some\nchanges in the code. So in the future\nthe job of software engineering the job\nof AI developers sitting in the audience\nwill probably evolve into being um agent\norchestrators right agent managers that\nare able to go deep. That being said, I\ndo see uh a lot of undertaking of other\nresponsibilities that have to do with\nbuilding software, which is there's a\nlot of product management. I see a lot\nof engineers would have to learn a lot\nabout um product management and actually\nspeaking to customers, they would have\nto learn design. Of course, you can get\nthe AI to design your front end for you,\nbut design that actually works would not\nbe found by vibes. They're found by\nspeaking to customers. So developers\nwill have to start speaking to\ncustomers. So we have this role\nflattening that is going to happen where\nwe see a lot of engineers becoming\nproduct managers becoming design and\nthey understand how to speak to\ncustomers in UX and UI. So there's this\nflattening that is going to happen and\nthat's what the future looks like. It\nlooks like generalist\nbeing able to understand what the agent\nis creating and be able to deep dive but\nalso have crossf functional skills that\nhave everything to do with product\ndevelopment. I don't know what do you\nguys think?\nI think you all inspire me to make a\nprediction and I'm not going to be\npaying the price for that in the future\nbecause predicting about AI in front of\nan audience of AI experts is the\nriskiest thing I will ever do. Uh but I\nI do think that manager as a badge of\nhonor in your job title will soon\ndisappear and the reason is is becoming\ntable stakes. Either if you manage\npeople or if you manage agents\nultimately every single person will have\nto gain those skills faster than we ever\nexpected. And the reason being even if\nyou're a junior engineer today you're\nexpected to manage multiple agents. And\nwhat's going to change ultimately is\nthat every single person will become an\nowner of a problem. If you think about\nit, if if you're a manager today, what's\nexpected of you is to own a problem, own\na surface, make sure that that project\nactually happens, this is going to\nhappen at the level of the individual\ncontributors. So as much as you know, I\nlove to say in public that everyone is\nbecoming a manager of agents, I do\nbelieve that manager is going to\ndisappear. You know, each one of us will\nwork on big problems and we're going to\nbe have to be proactive and run as many\nagents as we need using as many tools as\nit takes in order to get the job done as\nfast as possible. That's the future and\nI think it's going to happen way earlier\nthan everyone expects.\nSo basically the job disappears but the\nskill of managing still stays but with\nevery\neveryone\neveryone. Okay. when it comes to let\nlet's talk about this one skill that\neveryone should be learning because I\nguess there are a lot of people in the\naudience right now who are thinking okay\nI need to take the next step in my\ncareer from your perspective what is the\none skill they should be mastering right\nnow and the most important question how\nI think Dan had a good point with uh\ncritical thinking I mean you brought the\nnotion of debugging for example right\nand how monstrous that was if you didn't\nknow how to debug critically how would\nyou know what to debug or even if you\nneeded to in the first place you'd say\nwell looks great let's move on so I\nthink that that's maybe the one thing\nbut that that also I think you have to\ndevelop a sense of taste as well right\nand so this is\nlove that\nbut it's hard for juniors right if\nyou're a junior right now like how do\nyou get the critical thinking and that\nsense of taste because taste comes\nthrough experience taste comes through\nknowing what is great and what sucks\nterribly right but that only comes\nthrough knowing and seeing enough\nexamples and so um again I I don't\nreally have an answer except that those\nare the things I think you need to\ncultivate, but how you get there right\nnow. There's not really a deep learning\nclass on taste development. Maybe there\nshould be.\nNo, I think I think it's great that you\ntouched upon taste. I was talking to\nBill Gurley who's a legendary investor\nbenchmark. He actually told me software\nengineering is going to die really soon.\nI'm like I do not agree, but that was\nhis thing. He was like anything\nconnected with languages, learning\nlanguages or using a language talk to\ncomputer. But what he said was taste\nthat it's going to get really important.\nAnd the way you master it is you find\nsomeone either on social media, ideally\nin real life and just follow them and\nlook at their decision making. Learn\nwhat they think is good and what's bad\nand this is how you develop it. But I do\nuh do agree with you especially when it\ncomes to me hiring people these days. So\nmany things that I do can be automated.\nBut how do I trust someone make a\ndecision about what AI\nenhanced or created content gets posted?\nThat person has to have taste.\nThat's it. I'd say the other thing too\nand what Michelle was talking about with\neveryone's going to be a manager of\nsomething that's not human per se. Um\nunless you're a construction crew or\nsomething, then you should probably have\nmanagers. Um, the point is it's it's\nalso how do you work how do you manage\nyour time and work autonomously in a way\nthat's going to be effective. I think a\nlot of people are probably trying to uh\nspeedrun token max whatever and you're\ncreating more code but that's not really\nthe same thing as building something\nthat's useful right you're just cranking\nout more uh lines of code and burning\ntokens and so I think it's also like how\ndo you manage yourself especially when\nyou are managing basically your work and\nthe work of um of agents and I think\nthis is an unexplored area but you're\ngoing to have to get really good at that\num and I because I know that for every\ntask I complete with AI 10 more tasks\nopen up and the temptation is well let's\ndo those too, but then you have to pick\nand choose, right? And so I think this\nis increasingly going to be sort of we\nbasically the world we're in right now,\nyou can just basically invert it. And I\nthink that's sort of uh the the the\nworld you're about to enter.\nSo we're basically all becoming\nentrepreneurs have to manage their own\ntime and also have great taste,\nright? Have good taste and deal with a\nsituation of abundance, right? This is a\nthe paradox.\nDan, what do you think? Yeah, you know,\nagain, I we talked a little bit about\nthe skills, the critical thinking piece,\nbut I I think one thing that's much more\nprevalent today than was kind of when I\nwas going kind of through the the the\nclassic traditional education system. I\nmean, of course, we're here at the deep\nlearning event. Uh you know, Andrew with\nwith Corsera and these other aspects. I\nthink we talked to a couple you even\nsaid you had some education company that\nyou had done. You told me in the back. I\nthink what what's probably the most\nimportant skill if I were to to flip it\nhere because there's so much content\nthat's available because there is um you\nknow as you even mentioned uh you know\ninfluencers that you might um you know\nwant to follow in a certain area that\nyou want to get into. I do think that\nyou know continuous learning even\nwhether you're young or even as you're\nyou know older in your career people\nthat have that critical skill and and\nthat that desire um to you know continue\nto uh jump into the latest and greatest\nit is all moving so fast but that's\nwhere I think now you don't necessarily\nhave to go to you by the way I'm not\nsaying not go to uh you know your go get\nyour undergrad or masters in these\npieces that's not not my dynamic but I\nthink there's a a complimentary element\nwhere you can actually get the skills\nyou're looking for, get in deep,\nexperiment. Um there's so many tools\nthat are everything from open source or\nyou can get in going from a free trial\nof very low cost. So things that you\nknow 10 20 years ago would have been\nunimaginable. So I think if if you have\nthat uh critical thinking, you know,\nthat's a critical dynamic. But I also\nthink that uh ability for that\ncontinuous learning, that skill to jump\ninto areas to solve problems and then\nyeah experiment with it. You know, you\nhave things like replet where you can\nvibe code. You don't have to necessarily\nalways take everything into we think\nabout taking things into production for\nenterprise customers. But if you as a\nbuilder are experimenting with area, you\nstart understanding what people are\ntalking about and how things are\ndifferent in the software development\nlife cycle and how you do things today\nversus you know the past 20 30 40 years.\nSo um I think that would be what I might\nhighlight as well.\nYeah, love that. Every day at least like\n30 to an hour 30 minutes to an hour of\nlearning trying to play with the new\ntools. Richmond, what do you think? Do\nyou know what? Damn, you took my answer.\nI [laughter] love it. We're in sync.\nYeah, we're in sync. And I've had a a\ndecent career. And one thing that has\nstayed constant is just continuous\nlearning, right? Having that growth\nmindset is going to be the most\nimportant skill. So you've said it, I'll\nprobably double click into it and I'll\nadd another answer to it, which is\ncontinuous learning, always learning.\nLearning to learn is also a very\nimportant skill for every developer in\nthis room. This space is moving so\nquickly and anyone that tells you that\nthey're keeping up with the space is\nlying. No one is keeping up with this\nspace, right? The progress that are\nhappening at such an exponential rate,\nit's just impossible for one person to.\nBut if you're able to learn how to learn\neffectively, you'll be able to get\ninformation very quickly, digest it. And\nthis is another important skill you guys\nhave to cultivate is communicating that\ninformation that you're learning and\ncommunicating in different ways. It\ncould be communicating sitting on a\npanel like this or writing a LinkedIn\npost or writing articles or writing a\nskills.md file for your agents.\nCommunicating what you're learning in\nyour job process in your exploration.\nAnd if you're able to do this thing\nwhich is continuous learning and\ncommunicating what you're learning you\nwill really go far in your career in\nyour jobs in what you're trying to do\nfor yourselves or your customers. Then\nthe same quite in in parallel the same\napplies to LLM and agents as well. So\none thing that we're seeing in Oracle is\ncontinuous learning for agents is also\nvery important. So we have to\ncontinually learn and so do the LLMs,\nright? We have to have this continuous\nlearning that we never stop growing both\nthe agents and the humans that are\nbuilding this agent. So I think you hit\nthe nail on the head just continuous\nlearning.\nWhat do you think?\nYou probably heard of this concept\ncalled the uncanny valley especially on\nvirtual realities was this period of\ntime where things look good enough but\nnot extremely good to the extent that it\nwas almost repulsive. It was a bad\nexperience to use those products. I\nthink we're in the same phase right now\nwith coding agents, especially with\ncoding agents. The idea that they're\nreliable, they're very powerful. You\nnever know when they're actually lying\nto you or they're twisting part of the\nreality. The most important skill that I\nbelieve we all need, especially today,\nis truth seeking. You need to be able to\ngo deep and find the truth in case you\ndon't trust the results you're getting\nfrom an agent. And it's easy to become\nlaz in this period of time because a lot\nof work is delegated to an AI system and\nthe moment it becomes say more than 90%\nreliable your mindset is to become\ncomplacent. You just trust all the\ninformation that is thrown at you and\nthen you make decisions in your company\nbased on what your agent told you. So as\nespecially as a manager as a leader in\nyour company what's going to become even\nmore important is to ask yourself is\nwhat I'm reading correct or not? And if\nthat's not the case, the best thing that\nyou can do is actually to keep digging\nwith your AI because no one is going to\njudge you if your questions are stupid.\nIf you're in a meeting and you need to\nface like another leader and you're\nquestioning whatever they're bringing\nthere, it requires some level of\nfinesse, you know, human touch. Uh but\nwhen it comes to asking your agent, you\ncan keep hammering their systems until\nsometimes they literally contradict\nthemselves in the same session. And\nthat's when you know that the data\nyou're in front of is not correct or\nwhat you build is not perfect and you\nkeep seeking the truth until you know\nyou build what you actually need. It's a\nvery rare skill that I see maybe even\nless and less apply today because we we\ntend to lean too much to AI.\nSo for people who are listening to this\nthey're like okay so we're talking about\nbecoming generalists continuous learning\nunderstanding what is good what is best\nbasically developing taste. So what\nmatters most deep technical skills or\ngeneralist understanding of the real\nworld domain you're working in?\nI would say it's definitely kind of the\nmaybe a combination of the generalist\nand uh um domain expertise. Uh I mean\nthe tools are changing at such a rapid\npace right now that I don't know you can\ncertainly go deep on technologies. I\nthink it helps. I'll give you a really\ngood example actually. My my friend Wes\nMcKenna wrote this thing called pandas.\nHas anyone used pandas before? um right\neverybody. Uh so he he's uh he was sort\nof an AI um uh skeptic I think in uh\nmaybe last year or something but then he\nfound clawed code. He's like this is\npretty cool. So now he now he's actually\nconvinced and I I would actually\nattribute the rise of Python and in no\nsmall part to what Wes did with pandas.\nUm I think it it you know Python was a\nkind of a nobody language and now it's\nthe top language in the world. That\naside, he's actually moved on to Go,\nwhich is interesting, right? And he and\nhe says Go is the best language for\nagents. Here's the funny thing. Wes\ndoesn't know Go at all, actually. Right.\nBut he knows what good software looks\nlike. And so he he's he's coding all\nthese um new tools in Go, a language\nthat he doesn't know. But is he a deep\nengineer? Yes, I would I would consider\nhim one of the best engineers in the\nplanet, period. Um, and so I think\nthat's a good example of uh, you know,\nin the old in the old days like three\nyears ago, um, you would have had to\nbecome a go expert and that would have\ntaken you probably three years to become\na go expert. Now you can write\nproduction grade go and not even know\nanything about it. Is that dangerous?\nMaybe. I don't know. Could be. Um, but\nthe thing is that's what he's doing and\nis having a great success and I think\nmore importantly great fun with it. And\nso I think that's that's kind of where\nwe are now. But but again he was a very\ngood engineer and knew the domain that\nhe was trying to solve. And I think\nthat's uh but that points to maybe the\nlatter case being more evident. But\nDan\nyeah um you know it's it's interesting\nyou said is it is it more important to\nbe generalist or or kind of deep\ntechnical um\nI'll say you know this sounds almost\nlike a consulting answer but but to me\nit's almost like a blend. I think kind\nof classically when you looked at\ncompanies in the past, the number of\npeople that were in companies, um, you\nknow, like how can you start a company\nnow with so few people and and do so\nmuch? Um, I I think it's because you can\nwhether you're a generalist or you have\ndeep technical skills, you can actually\nuse AI to kind of fill in the gaps in\nthe areas you don't know. Um, sometimes\npeople fill in the gaps and maybe they\nshouldn't. Sometimes I hear, you know,\num, oh, don't use your LLM for medical\nadvice or even though, by the way, I\nthink it does a really good job the\nmajority of the time. And I've actually\nused it a lot to help with like my\nparents and other aspects. Sometimes my\nkids just call, you know, calling up\nfrom college, something's going wrong,\nthey don't feel what. So, you know, jump\ninto that area. But I I think the reason\nI bring that up is whether you're deep\ntechnical um, and you need to fill in\nsome of the generalist areas that you\ndon't know, that's more easily done. And\nI think if you're more of a generalist\nand you're trying to understand, you\nknow, whether it's trying to build\nproduct or these other aspects, even how\nwe talked, I think we've talked a lot\nabout like engineering and product\nmanagement kind of blending. But I kind\nof see, you know, I I couldn't tell you,\nyou know, maybe I would lean 6040 now or\nor maybe slightly above that toward\nhaving like a if I'm starting a new\ncompany, maybe the CEO being more\ntechnical. Um, but I I'd like a blend of\nthat dynamic. So I don't know that I\nwould pick one more important. I I would\nsay that the great thing is now because\nof AI I think wherever your gaps are you\ncan fill that in very quickly whether\nit's domain expertise uh that you might\nneed that you're lacking or on the vice\nversa side if you're you know doing some\nvibe coding I mean I even think about\ninternally at my company you know Andrew\nis always pushing you know I even want\nthe field the the go to market team to\nknow how to develop and and and we did\nthat you know two years ago I remember\nputting together a thing for our our AES\nto here's you know very you know getting\nstarted with Python for beginners and\nthese things and I'd say 40% really gave\nit a try and cared. Now we just did a\nwhole new curriculum for our developers\nusing uh different agents, agent coders,\ndynamic, etc. And the same people that\ndidn't want to spend a second learning\nPython are creating some of the best,\nyou know, custom CRM applications to\nhelp them attack their pipeline or, you\nknow, I I remember having an issue with\nmy customer success leader. I hope Ron's\nnot here right now. You're doing a great\njob if you are, Ron. But a couple years\nago, I remember him buying like a SAS\napplication, taking forever to try to\nget it set up. Then we're like, you know\nwhat? Let's just get rid of it, etc.\nHe's now like in two weeks because of we\nhad our data in great shape. He's now\nbuilt an incredible tool that can show\nus the usage of every single one of our\ncustomers, who is above usage, below\nusage, critical alerts, you know, uh if\nwe're if they're getting errors in real\ntime. These are the same team members\nthat didn't want to touch that because\nthey were like code syntax that I just\nit wasn't in their DNA. Now now they're\nblended together, right? So I I think\nthat to me is is you know you're kind of\n50/50 both.\nSo it is everything is blending. Uh\nthat's definitely a trend that we're\nseeing. Richmond, what do you think?\nDan, you keep taking my answers. You go\nfirst. I should have sat over there.\n[laughter]\nWell, it's great we have consensus.\nYeah. Yeah. Um so I think the way the\ntraditional education system is kind of\nshaped is you have this you start off\nbeing a generalist then you kind of\nspecialize in like a specific field but\nthe benefit of AI today and the tools we\nhave one of the great benefit is we have\nthe ability to actually all become\ngeneralist very easily. Um, it could be\ntaking a PDF of some subject, turning\nthat into a podcast, and you listen to\nthat while you're in the gym, and then\nyou're conversation ready, right? That's\nwhat being a generalist means today.\nBeing conversation ready to hold um, uh,\ninteractions about a certain topic. But\nI do think tech deep technical expertise\ndoes have its place, especially in a\nfuture where it might become rare.\nI I'm changing my my my answer as we are\nspeaking. I agree with you. But I do\nthink there are some folks that if they\nfind the right deep technical expertise\nto specialize in, they will really\nprovide immense value in society or\nwhatever domain they're going in. And it\nwill be a few folks. It wouldn't be a\nlot of people. But um but the takeaway\nfrom this is\nit's easy to become a generalist today.\nYou might as well just do it, right? So\nit's not a it's not a should I be a deep\ntechnical expertise or have be a\ngeneralist. It's both. It's always both.\nI will echo your answers. I think\nwe should switch.\nYeah. Yeah. We should switch this again.\n[laughter]\nUm I think specialists will become even\nmore specialized and generalists will\nbecome even broader and the reason is\nthere is a class of hard technical\nproblems and instead of shrinking is\nactually expanding. If you're a big\nadopter of coding agents in your\ncompany, you know exactly what I'm\ntalking about. the type of technical\nchallenges we're facing on a daily basis\nin in including the volume of code that\nis been generated. What do we need to\ntest? How fast is the testing pipeline?\nAll these problems existed in the past.\nWe feel 10x the pressure that we felt\nliterally a few months ago. So very\nspecialized. So they has to move even\nfaster. But my recommendation is\nspecializations will not last anymore\ntwo decades. We'll have a short window\nof time maybe in a matter of like a few\nmonths or a few years. and you need to\nconstantly update your prior and become\na specialist of something else in order\nto be very relevant. Generalist is\neverything I was saying before. If\nanything is the most exciting time to be\na generalist because your bread expands\nby orders of magnitude just if you know\nhow to master AI tools. So great time. I\nwouldn't change your professional\ninclination. If you're an amazing\nspecialist, please keep being that and\nI'm going to be hiring you because\nthat's exactly what we need.\nYeah. It's also about understanding\nyourself, right? we're leaning towards.\nWe have one minute left and I have one\nvery specific question.\nSo, we talked a lot about skills and\neverything, but let's try and make a\nprediction. In five years, let's take a\nbusiness that is retail based. So,\nthey're not like a tech company. Today,\nthey probably have 10 developers to run\ntheir website. And there is a series A\nstartup that is hiring their development\nteam. So, can you give me a number for\nthat retail business? how many developer\nhow many developers they will need in\nfive years and a series A startup who's\nhiring their development team how many\npeople will they hire\num I would say that the the uh retailer\nif they kept the developers uh and they\nhave the same operating practices\nthey'll probably need no developers so\nthey might be out of business um and\nthen the series A um yeah I think it's\nTPD it might be might be the same number\nit might be a thousand I don't know so\nit's a good\nquestion okay yeah it's it's something\nto think about maybe based trends that\nyou're seeing.\nSo, so you know I I'll I'll say if\nyou're saying and let me just play it\nback. If if in retail today series A\ncompany has 10 developers, how many will\nthey have in the future in five years?\nIs that kind of the question?\nUm so two companies, one is retail\nbased. They have 10 developers to run a\nwebsite. Will they need them in five\nyears or not? And serious a tech\ncompany. Will they be hiring developers\nas much as they're hiring now or they'll\nbe hiring less?\nYeah. So, good question. So, so first of\nall, I I don't think they'll need as\nmany developers at series, like in the\nretail dynamic. So, so in that dynamic,\nI think you will see the uh rise. You're\nalready starting to see that um less\ndevelopers being able to do more early\non. Um whether you're a mature company\nor or a series A company getting\nauthoriz,\n75% less, I I can't tell you that bad a\nnumber, but I'd put it in that arena. Um\nbut I I think what what will be if doing\nthings right you should be able to get\nto the series A um you know dynamics and\nmaybe series A of the future there's\nhigher levels of revenue that will need\nto be achieved because things are so\nmuch more efficient etc. But I I think\nyou you will still ultimately need to\nadd more and more developers as you kind\nof max out in in this area but I think\nin the future to get to that level of\nwhat you consider series A today versus\n5 years out both will be less.\nYeah, I think for me the answer is\nsimple. Both companies, [snorts] series\nA or the existing realtor company will\nprobably hire more. Now, would they be\ncalled developers is the question,\nright? Because when we say developers,\nwe're talking about people that are\nbuilding these softwares, right? And\nthey're building the building blocks of\ntheir softwares. The question is, would\nthat role still be called what it is\ntoday?\nAnd whatever it's called then they would\nhire\nthat.\nboth companies will hire even more\ndevelopers than before. They will have\ntotally different functions. I think in\na retail e-commerce, there will likely\nbe no one maintaining the e-commerce\nplatform because we will know exactly\nhow to run them. There are nice sus\nsolutions for that. A lot of functions\ntoday are automated by people that have\na developer mindset. We see this at\nRapid as well. You know, our go to\nmarketing,\nHR, a lot of people have a developer\nmindset. the stark coding agents to\ncreate internal tools to make that\nfunction much more lean. So every person\nin that company will actually be writing\nsoftware as a series A ventureback\ncompany you're shooting for the stars by\ndefinition that what investors want from\nyou. Of course everyone will be strongly\ntechnical there and they will be 100x\ndevelopers compared to what they are\ntoday because of coding agents but they\nwill actually be strong specialists.\nYeah. And uh to wrap up on a positive\nnote, I'm talking to a lot of\nentrepreneurs. As entrepreneur myself, I\nalso see how with one person, we're\nbecoming I don't know 2x more\nproductive, which makes me want to hire\nmore. So I think we all agree that\nsoftware engineering has a bright\nfuture, but it's going to be slightly\ndifferent with more people becoming\ngeneralists, owning [music] product, uh\nand developing taste.",
  "transcript_chars": 57672,
  "ingested_at": "2026-05-21T19:15:51.984888+00:00",
  "source": "retry-no-transcript",
  "yt_meta": {
    "view_count": 233,
    "like_count": 4,
    "channel_id": "UCcIXc5mJsHVYTZR1maL5l9w"
  }
}