{
  "video_id": "WOPIoZuD1og",
  "channel_slug": "techwithtim",
  "channel_handle": "Tech With Tim",
  "title": "The State of Software Engineering in 2026",
  "duration_seconds": 670,
  "url": "https://www.youtube.com/watch?v=WOPIoZuD1og",
  "upload_date": "20260430",
  "transcript": "In January\n2026, the head of Claude Code at Anthropic said\nhe hasn't written\na single line of code by hand in over two months.\nNow, across the whole company,\n70 to 90% of their code is completely AI written.\nMeanwhile, employment for developers\naged 22 to 25 is down nearly 20% since ChatGPT\nlaunched an entry level\njob, postings have dropped anywhere\nfrom 46 to 67%, depending on who's counting.\nNow, AI isn't coming for software engineering.\nIt has already arrived.\nSo here's what the state of software engineering\nactually looks like in 2026.\nAnd more importantly,\nwhat nobody's talking about now.\nWhat I'm going to share in this video\nisn't just a bunch of stats.\nThis is my personal experience of what I've noticed,\ntalking with hundreds of different developers\nthat are at the forefront of AI, that are seniors,\nthat are juniors across every single discipline.\nNow, what I've noticed,\nfirst of all, is that the role of the software\nengineer is shifting from a producer to a supervisor.\nNow, what that means is we're no longer really\nresponsible for writing massive amounts of code.\nAI code generation has gotten so good\neven in just the past few months,\nthat if you're not using AI,\nyou really are just falling behind\nand really not productive at all\ncompared to the developers who do utilize it.\nThis means that you do need to use AI tools.\nYou're going to AI generate a lot of your code.\nAnd what's more important is being able to understand\nwhat good actually looks like.\nYes, we can all generate massive amounts of code,\nbut being able to have that experience to understand\nwhat code is good, what code is bad,\nand what's going to be maintainable\nand just proper for\nthe future, is really the skill that matters more.\nNow, this is something that you can only develop\nwith a lot of experience, but we're in this kind\nof tricky place where in order to get the experience,\nyou really should be writing the code by hand.\nBut you can't write the code by hand,\nbecause then you're going to be too slow\nand you're not going to be productive\ncompared to the other engineers.\nThis will lead me to another point\nI want to talk about later in the video,\nbut it's a really weird time\nwhere in order to get the experience required\nto generate a bunch of productive\nAI code, you need to have written code manually.\nBut you just can't write code manually because\nthen you're so slow compared to everyone else.\nThis is a really interesting dilemma,\nand as someone who teaches a lot of coding concepts,\nI'm kind of in the middle here\non what I recommend to people learning.\nShould you purely just use AI?\nShould you actually write\nor learn the fundamentals from scratch?\nShould you do a combination?\nOf course, it's somewhere in between,\nbut it's a really tricky place,\nespecially for junior developers.\nAnd to be honest with you,\nI don't know what the right answer is,\nand I'd be curious to hear\nwhat you think in the comments down below.\nNow, the next major trend that I'm noticing\nis more related to the product side of things,\nbut that's something called Agent Experience or AX.\nAnd we used to have UX, right,\nwhich was user experience.\nBut now agents are more and more actually consuming\nthe software that we use on an everyday basis.\nFor example, I have like a Monday dashboard\nor an Asana dashboard or my Google Drive,\nand rarely now am I actually going in there\nand manually doing something.\nI'm purely communicating with an AI agent,\nthen takes the action for me.\nNow this is becoming a more and more important\ndesign discipline and a design decision\nthat teams have to consider.\nHow do I design my software in a way\nsuch that AI agents can consume it and use it?\nAnd we've seen that with MCP,\nwe're seeing that with plug ins, adapters\nconnecting to tools like Open Claw.\nAnd this is becoming\nmore and more popular in the industry.\nAnd something that I think is going to be a pretty\ninteresting kind of avenue to dive into or something\nwhere there could actually be a lot of jobs\nwhere your head of agent experience, right?\nOr you're an agent experience developer\nwho's really responsible for making sure\nthat this software is adaptable\nor usable with AI agents.\nAll of the major tools out\nthere are coming out with MCPs updating them quickly,\nand even having,\nyou know, automatic prompts that you could just copy\nand paste into your tool\nthat will set everything up for you.\nReally interesting trend.\nAgain, curious to know what you guys\nthink about this, but something I'm noticing a lot\nthat's been wrapping up\njust over the past few months.\nNow I have a few other more dramatic insights\nthat I want to share,\nbut quickly I need to give a shout out to today's\nsponsor, Nexus Dot AI.\nNow, if you're anything like me, you probably have\nfive different AI subscriptions open right now.\nChatGPT for one thing, Claude, for another Gemini.\nWhen you need long contacts\nplus whatever new model dropped last week.\nRight now\nit's a mess and you're paying for all of them.\nSomewhere between 60 and $200 a month.\nWhen you add them all up. Now next, AI fixes.\nThat is one platform\nthat gives you access to every major model like GPT,\nClaude, Gemini, all of them\nthrough a single interface and a single subscription.\nSo there's no more tab switching,\nno more re pasting the same context\ninto three different tools,\nand no more five different separate builds.\nNow the way I actually use it is pretty simple.\nI pick the model for the job and it just runs.\nSo writing or strategy.\nYou know, I'm using cloud research\nwith a ton of source material.\nI'm using Gemini.\nQuick answers I'm going to GPT now.\nJust takes two seconds to switch the model.\nAnd you can do it all in one place.\nNow the feature that I keep coming back to\nis deep research.\nYou give it a topic, it goes out and pulls\nfrom the web and hands you back a fully cited report.\nWhat used to take me two hours of googling,\nreading and compiling now\ntakes me about three minutes\njust waiting for the AI to respond.\nNow, it's currently replaced at least three of the\nAI tools that I had in my stack.\nNow they also have no code AI agents\nwhere you can describe a task\nin plain language,\nand it builds an agent that does it on repeat.\nThey also have zero training on your data and are\nfully compliant with things like Soc2, GDPR and more.\nNow people are saving up to $200 a month\nby switching to next Thursday.\nAI because the one plan replaces everything\nthey were paying for separately.\nAnd right now you can try it for free for seven days.\nI'll leave the link in the description, and you can\nuse the code Twit for 10% off after your trial.\nNow it takes two minutes to set up.\nIt's genuinely worth checking out.\nNow let's get back to the video.\nNow, the next major thing that I'm noticing is that\nit feels like we're slowly losing engineering talent.\nNow, what I mean by that is, even in\njust the past few months,\nwe've seen all kinds of breaches, hacks, exploits.\nIt seems like software is getting worse\nand worse quality and compound that with the fact\nthat less and less junior devs are getting hired\nand getting trained, and more and more\nseniors are retiring, as always happens.\nWe're kind of in this weird position\nwhere in order to be useful to a company.\nYou need to be a senior engineer.\nSo these companies are not really hiring juniors\nthat much.\nAs we've seen, it's really difficult\nin the job market, especially as a new grad.\nBut then when the seniors start leaving,\nthere's nobody to replace them.\nNow this hasn't become a massive problem yet,\nbut obviously it's something that I think\nis going to compound over time where these companies\naren't investing in the younger talent,\nthey're not losing the money\nto eventually make it back in the future.\nAnd as these talented developers leave this company,\nswitch to another one, retire, whatever it may be.\nWe don't have any replacement.\nNow, this is also compounded with the fact that now\nno one is really learning or writing code anymore.\nThey're just going to an AI prompt right there\nsaying, hey, this is what I want you to do.\nIt's generating a bunch of code.\nThey're quickly reviewing it\nand then they're pushing it into production.\nThis means that we no longer having all of this\ninnovation, all of these new solutions that we had\nbefore where people actually had to sit there\nand deeply think about problems.\nInstead, we're just really quickly\nspitting out a ton of alarm generated code\nand really just reusing the same code\nthat we've had for years.\nNow, this has always been the major problem\nwith labs.\nThey're good at predicting text.\nThey do that based on their training data,\nwhich means\nthey can only do something\nthat's already been done before.\nThey're not coming up with novel ideas.\nThey're not creating new inventions. Right?\nThey're not coming up with a brand new framework\nthat's completely changing the industry.\nThey're just using what already exists.\nAnd unfortunately,\nmost of what already exists is really bad code.\nAnd now we're seeing that being forced\ninto all of these products\nbecause of how much lip generated slop\nis just being pushed out into the market.\nNow, don't get me wrong\nhere, LLMs are absolutely necessary.\nNow we have to use\nAI if we want to keep up with the way\nin which a lot of developers and companies are using,\nit can be more harmful than good and lead\nto a lot of really difficult to maintain.\nAnd just, you know, strangely written software.\nAnd I think we're seeing that more and more and,\nand I don't know, at least in my own perspective,\nit feels like there's just like a new hack\nevery day, a new issue, a new exploit.\nAnd when you have AI writing its own software,\nthere's all kinds of weird bugs and problems\nthat pop up\nthat are just really difficult for humans to solve.\nAnd this is going to become, in my opinion,\na bigger and bigger issue\nas more and more code gets AI generated.\nNow, the last major thing that I've noticed over\nthe past few months here\nis that the bottlenecks related to software\ndevelopment have moved.\nNow, previously, the bottleneck to developing\nsoftware was writing the code.\nIt was very expensive to have,\nyou know, a team of hundreds of engineers\nsitting there producing massive amounts of code.\nAnd it took, you know, a non-trivial amount of time.\nWe've now moved to the phase where code is cheap,\nbut everything else isn't.\nSo the scarce skills are things like system\ndesign, operational excellence.\nRight.\nSo observability, incident response,\nyou know, debugging in production and judgment\nabout trade offs.\nNow things like front end have had massive\nproductivity gains because that's really where\nit's just writing a lot of code.\nSure, there's a little bit of thinking,\narchitecture and design, but it's not as critical\nas something like backend development\ninfrastructure, DevOps deployment.\nSo now the bottleneck is not really\nhow much code can we write?\nIt's how much code can we review, ensure is secure\nand deploy maintain in production\nwith all of the features that we need.\nNow this is really shifted the demand\nand need in the market for senior engineers\nthat have those skills already,\nwho have dealt with these issues in production\nand have the ability to orchestrate and manage\nmultiple things at once.\nAnd that's why we've seen less demand\nfor junior developers, because we don't just need\na bunch of random features written anymore,\nwhich we typically delegate to them.\nWe need that higher leverage task to be completed.\nSo that means if you're someone who's trying\nto get into this market, as much as it's not easy\nor something you want to hear, these are really\nthe skills that you need to focus on.\nYes, you should still be able to write code,\nbut it's not about how much code you produce now.\nIt's about how good that code actually is.\nAnd again, those higher level skills\nthat allow you to develop a scalable system\nthat's actually going to be profitable\nand usable for a company.\nNow, I am genuinely very scared about what\nthe future of software development\nis going to look like because of how much\na high charter rated stuff we have,\nand because of that loss of talent\nthat I talked about before.\nAll I can say is that keeping up to date,\nstaying adaptable, keeping an open\nmind is more and more important,\nespecially today, than it's ever been before.\nAnd if you're a developer who's refusing to adapt,\nyou will be left behind.\nAnd what we want to happen versus\nwhat needs to happen are two very different things.\nSo as much as you might not love the new market\nthat we're moving into,\nif you want to stay employable,\nyou do need to stay up to date.\nYou need to adapt.\nAnd that's really the only thing that you can do\nto make sure that you are going to be\nsafe, still have a job, and hopefully still be able\nto provide for yourself and your family.\nNow, look,\nI don't want to just be doom and gloom here.\nSome of these things are incredibly positive.\nWe're able to build much more\nthan we ever could before.\nI think it is still an exciting time right now,\nbut obviously it is uncertain.\nIt is scary and we just genuinely don't know\nwhat's going to happen next.\nSo we need to stay kind of\non the ball, adaptable, like I said,\nand just be ready for whatever occurs next.\nAnyways, guys, with that said,\nthat's all I had for you.\nIf you enjoyed the video, make sure to like,\nsubscribe and I will see you in the next one.",
  "transcript_chars": 13334,
  "ingested_at": "2026-05-21T19:03:23.584696+00:00",
  "source": "retry-no-transcript",
  "yt_meta": {
    "view_count": 50359,
    "like_count": 1211,
    "channel_id": "UC4JX40jDee_tINbkjycV4Sg"
  }
}