AI Coding is here... What’s next? My 2025 GenAI BIG BETS (AI Agents, o3, Paic)

what's up Engineers welcome back Indy Deb danan here welcome to a special video we're walking through my 2025 generative AI predictions we're going to cover my bets on AI coding large language models and generative AI companies throughout 2025 I'm going to share my secret as to how I stay ahead of the curve in the age of generative AI so that you can understand how you can get ahead of the curve as well here's the outline for this video here's what we're going to cover I want to talk about why it's important to make predictions and

make bets I do this every single year and this process is critical for My Success back to back year after year after year we're going to then Dive Right In to my 2025 AI predictions we're then going to discuss everything that's happened with principled AI coding huge shout out to everyone that has taken principled AI coding that is learning that is growing and accelerating their engineering with AI coding and then we're going to quickly review my 2024 performance as you can see we have a win rate of 87% it was a crazy accurate year I

made 41 bets and I got 36 correct and then lastly we're going to talk about how we can take these bets take our predictions turn them into actions which you can then get concrete results right what's the point if there's no return on investment on your predictions on your bets and on your actions so if these topics interest you stick around and and let's Dive Right In with all new technology the most important Advantage is time information and access you want to be the first to know about new technologies for the longest period of time

with the best access to them right information time and access so why is this this is because by being there early by understanding the technology and by having access to it you obtain outsize return on your time and energy before anyone else even gets the chance to imagine having access to 03 right now think of all the incredible things you could do with that technology so then the next natural question arises right what's the best way to position yourself to know about new technology early and have the best access to the technology so we're just

thinking through this process step by step okay so we have four options and we just going to walk through these one by one you can follow the crowd this is very clearly and very obviously the slowest way to position yourself to know about new tech early and have access okay if you're just following what everyone else is doing you are showing up very likely in the back of the pack right with the masses we can do much better we have three other options you can follow bleeding edge researchers Engineers Builders and companies okay this is

much faster right when you're following the engineers that are putting out the Next Generation language model technology AI coding tools coding patterns repositories and generally information you're going to have Fast Access okay but we can do even better what's faster than following the people releasing the new tools the new ideas the new technology the new models you can quite literally be the bleeding edge researcher engineer Builder company okay this gives you instantaneous access to know about the new tech early and have the best access okay so I hope you can see where this is going

right why do we make predictions why do we make bets we have one additional option here and that is of course you can predict what will come and bet on it with your time money and energy before it happens right this is beyond instantaneous there is no faster way to be in a position to win with a new technology than before it even exists okay and this is why it's important to plan to predict and then put your money time and energy behind your predictions right that's what changes the prediction into a bet when you

have some skin in the game this is why it's so important to make predictions this is why every single year on the channel I share my predictions for the year here with you because when you make predictions and bet on them you gain the earliest access for the longest period of time to the new technology information and opportunities before they even exist right before the bleeding edge researcher engineer or Builder has even created the thing so what is a concrete example of turning a prediction into a bet and then acting on it imagine you predict

that in 2025 we're going to see a new wave of language models that can support 2 to 5 million input tokens before before that even happens right before the bleeding edge researcher releases it before everyone that follows them sees it and before the crowd sees it you are choosing option four you're predicting that this will happen and you're betting on it by taking the time and the resources and your energy and you're building out this new context management system or a personal knowledge base where you can quickly build context relate context tag context and build

up a graph of information for your code bces uh projects documentation and large token sets that you can then combine into you know very large multi-million input token prompts right and you can do all this before it's even released okay so this is a concrete example of why it's important to make predictions and make bets this is how I stay ahead of the curve don't wait for things to be created make a bet assume it will be created and and if you're wrong pivot and if you're right you're there before anyone else okay let's move

on so couple prediction disclaimers and then let's talk about the themes for the predictions just to get this out of the way these predictions and bets are my own be skeptical form your own opinions don't blindly follow me or anyone without first forming your own opinions okay any prediction against or for a company is not a slight against them um it's just a prediction so I'm not here to you know bad mouth any company or llm or tool these are just predictions okay and then lastly I'm not sponsored by any company I'm not promoting anyone's

products or language models I'm not paid to make any of these predictions okay I'm a free agent third party one of the reasons why I want to stay this way is so that you can trust me throughout every single video that I put out and you can know that the only thing I'm selling is something that I am building for you we're going to talk about principal AI coding a little bit more further on in this video okay so let's talk about them so what are the themes for my predictions we're going to be talking

about AI coding large Lage model plus so you know AI agents personal AI assistants AG gentics geni companies and and a couple predictions kind of mixed in on the engineering job market okay this year I'm dialing in on fewer higher conviction predictions for 2025 we have a total of just 15 predictions down from I think I made 46 last year um I'm really dialing in I'm specializing I want to make higher conviction bets and actions against my predictions so just 15 predictions this year let's dive in starting with kind of a hot spicy take here

if you've been AI coding you probably don't think this is a spicy take but AI coding becomes the standard so this is how this is going to work I have a kind of headline I have a prediction and then I have a win condition and a conviction level if you're not using a to code you're not coding and you're not getting hired so here's the wind condition I'm adding the wind condition this year to make it a lot clearer for the end of the year whether I got the prediction right or not so this year

in order for this prediction to be right for 2025 we need to see a clear sentiment shift in engineering media from you know writing code with AI is unreliable unsafe so on and so forth you've already heard all this crap to you can't keep up with the pace of development without AI this is going to be the shift we're going to see and I have high conviction here so conviction levels low medium high I'm betting big when my conviction is high here AI coding becomes a new standard this is part of the reason I put

out principal AI coding more on that later all right so what's next agentic coding begins AI software Engineers become usable thanks to top tier reasoning models improved tool calling performance and more intelligence per input token and the win condition here is that you know Engineers that are on the edge start using agentic coding tools for simp simple tasks I'm not saying it's going to take over I'm not saying it's going to dominate the industry if you progress to the end of principled AI coding you know that this is possible you know that you can build

agentic coding patterns like this to solve you know small to medium difficulty tasks so the wind condition here is that we see a cheaper highquality Devon likee tool come out that takes the Gen AI ecosystem by storm and once again everyone's going to freak out AI is taking Engineers jobs uh that won't be true you and I both know AI doesn't take jobs Engineers using AI take jobs here's kind of a simple AI autonomy graph that I built out about halfway through the year I'll link that video in the description if you want to dive

in to that a little bit more but tier one we had co-pilots tier two we have ai coding assistants everyone is using these right now they they've essentially become the standard right think about cursor AER wind Surfer all these applications are kind of just all these tier 2s and then we started to see tier 3s in 2024 they were all really bad let's be real even Devon the best of them was very very limited but I'm predicting in 20125 a gentic coding begins all right let's move on so Evergreen AI coding experience we're going to

see a nonv code based non- agentic Bottoms Up Evergreen reinvented application for AI coding so this is going to be a completely new way to write code with AI beyond the typical chat window uh command control K experiences okay pretty much every tool I have listed here is more or less uh the same thing wither being a special exception here these tools share a lot of the same experiences right uh chat window codebase lookup command K selection prompts uh multifile editing I'm really looking forward to and I'm predicting a brand new revolutionary AI coding tool

that really changes how developers interact with AI for coding and to be clear I think it's going to be an AI code tool not an agentic coding tool so it's not going to be a Devon right I'm not talking about an endtoend solution I think we're going to see something brand new pop up with new pattern new UI new ux and I think it's going to be a really fantastic Evergreen AI coding experience it's going to force us to rethink not in just engineering but cross domains how we can really use language models I have

medium conviction here this is kind of a riskier one but I am Bing on this happening in 2025 the cost of code declines not experience so I have high conviction here and I imagine you can kind of already see where this is going right the cost of code continues to decline while AI coding becomes the standard right and then eventually agentic coding the cost of expertise experience great judgment and great engineering will continue to rise I do think that unfortunately we're going to continue to see the bottom of the job market kind of continue to

fall out right it's going to be harder for junior and mid and Noob Engineers to find their kind of first gig and get that initial experience so so this puts even more pressure on people that can do the hard work get the real experience and you know people that already have it if you're mid senior plus and you already have the experience you can now multiply it with AI coding tools and this is a big theme that I I reiterate several times in principled AI coding it's not that you can't get great engineering experience as

a noob you absolutely can but senior plus Engineers are uniquely positioned to utilize generative AI as long as they are willing to learn new tricks and change the way they think about engineering you want to be thinking more about what you want done and less how it's done hard win condition here Tech job opportunities is going to show clear signals that coding is not really the key part of the job because the cost of code is going to zero it's really going to be about the specialized expertise experience judgment and then the kind of raw

engineering skills right architecture modeling documentation that's going to be the key part of jobs and job descriptions okay not the coding so let's move on to our language model predictions so I'm really excited about these I have firmly believed as soon as I started seeing these articles pop up that AI is hit a wall there's no more data my immediate thought was you are vastly underestimating the explosive potential of not just the tools but the researchers and Builders behind generative AI there's probably like a 100 or 200 Engineers absolutely cracked engineers at these companies that

are essentially Paving the future of the tech industry as a whole and I think people just really don't think about that like these guys started popping off back in the 2010s and they haven't stopped and they're not going to stop you know just a reminder like you might see a logo but what's really Behind These great teams are key leaders and then cracked researchers and Engineers grinding because they love to do it okay there is no wall okay never has been uh I have high conviction here language models will continue to improve across all benchmarks

they're going to completely saturate various benchmarks we're starting to see glimpses of this with 03 there is no wall there never was the wi condition here is simple we're going to see three new state-of-the-art base models in 2025 and we're going to see a reasoning model it's going to hit 100% on a popular Benchmark okay I have high conviction here there is no wall there is more data they synthetic data and there are just entire different architectures paradigms and ways to build out language models to completely circumvent any pre-existing notion of a wall I don't

buy it never have this is something I'm really excited about we're going to see hyp specialized large language models in 2025 what do I mean by this I think we're going to see small single purposed and not even small it doesn't have to be small but we're going to see single purpose large language models emerge as AI agents as a service deep seek version 3 is a great example of this right this is a high- performing AI coding specific llm I think we're going to see this trend continue we're going to see entire products built

around single use case language models and AI agents and the term AI agents here is a bit more fitting because this does not need to just be a language model API call but it does need to be a specialized tool that you know ends up calling a language model at some point the win condition here I think we're going to see a launch of a successful product built around a single open AI compatible endpoint running a specialized AI agent for a specific use case I'm really excited about this I think this is a very powerful

idea I do get the sense that there are a couple you know small companies and maybe even big companies starting to chase this idea um but we really need to see them kind of dig in in order for this to to Really Happen we're going to be covering deep seek V3 in upcoming videos Sub comment like stay tuned for that so what's next so I predict we're going to see the year of infinite memory and not exactly infinite but it's going to feel infinite we're going to see context Windows expand dramatically to 1 to 5

million tokens inside of powerful base models the win condition is simple we're going to see a 5 million token context window base model I have medium conviction here we already know that uh Google is pretty close to this so I use this as an example of something you can concretely prepare for ahead of time to really maximize the value you can get out of this I will say though I do not think most Engineers most product Builders are utilizing the existing context window this is why my conviction is medium here I I really don't see

as many people really using the full potential of the Contex window as it really could be myself included only recently have I hit the 500k input token mark for a real kind of you know valuable use case I think we're still going to see this I think that you know engineers and Builders pushing the input token window do want to see a larger context window all right so next prediction I think we're going to see a GPT 40 level slm what's an slm small language model we saw 54 come out a couple weeks ago still

waiting for the official official release from Microsoft this has been really really incredible I got my hands on a version of this this model is insane and it is very very tiny so the prediction here is simple we're going to have GPT 40 LEL small language models on device the win condition here is that we're going to see this embedded inside of the operating system undetectable without benchmarking you know to be super super clear here I mean consumer Hardware truly consumer Hardware not developer Hardware I think we're going to see this you know on truly

consumer facing devices we are seeing early signs of this Mac OS devices are going to have a you know on device language model new windows devices are getting this I think this is going to be a continued Trend I'm very confident we're going to see DPT 40 level small language model models so this is an easy prediction to make but I think the amount of it that we're going to see is not truly understood we're going to see exponential slop gener of AI unleashes infinite content repurposing and generation and it's going to flood the internet

okay it's going to make it really hard to identify valuable signals and legitimate information from text to images to video we're going to see exponential slop on the internet and so I think as a win condition here we're going to see big platforms put up more content detectors so more AI content detectors but also require new flagging of AI generated content on social platforms High conviction here I don't think this is a difficult prediction to make but concretely making this prediction puts me in a place puts myself and puts you in a place to really

think about how to keep track of valuable information sources and valuable information platforms that are rich in actionable information not just hype arousal and you know clickbait you don't want those you do want actionable information okay so we're going to see exponential slop so let's turn our attention to generative AI companies I'm making this buold claim here I made this prediction last year I was right no one caught up everyone thinks everyone caught up but then we saw 01 we saw 01 Pro and now we've seen 03 open AI is going to remain number one

they're going to remain the top llm player with the best class most intelligent model Google is heating up anthropic is competitive right Cloud 3.5 Sonet is a top tier model and llama is quite literally killing the market by putting out you know free open source models llama 4 is right around the corner as we know but in the C there is only one king and it is open AI they're going to maintain their lead I do think everyone is continuously underestimating how far ahead they are how big their lead is even with all the exits

some of the researchers leaving some of the key leaders leaving I do not think that they have lost their lead they're going to need to make a massive blunder and I don't see a 12-day back-to-back release ending in the release of a Next Generation Benchmark saturating reasoning model a mistake okay I just don't see them on the course of making a mistake open AI remains number one that doesn't mean Google and anthropic aren't producing high value important you know useful models same with metas I just don't see them taking the throne from open AI in

2025 here's what we're going to see three independent benchmarks showing an open a model at the top across 75% of the benchmarks okay so that's how we're going to denote them being the best I have high conviction here open AI is in the lead and they remain in the lead big Tech shrinks smbs grow this is not just any smv though so this is small to medium AI powered businesses Okay so so big tech companies continue to reduce their Workforce just to be super clear reduce their Workforce not their influence not their capabilities not their

market cap okay they will continue to grow these big tech companies especially these you know gen tech companies they are awake they are driving fast they are making plays Google is the biggest example of that the sleeping giant has awoken at the same time though we're going to see small businesses grow and grow and grow thanks to the productivity of AI tools so the win condition here is simple we're going to see three big tech companies decrease Workforce by 10% in 2025 and we're going to see three small businesses with fewer than 50 employees reach

$50 million in valuation okay so medium conviction here feeling pretty good about this one but also not putting too many coins in this there not a lot of actionable information here uh to actually act on data greater than ux greater than LM bench greater than everything else what is this prediction what do I mean by this so a business's most valuable asset become their data user experience and specialized llm benchmarks this is happening while AI tools basically fully commoditize every other piece of the business okay replicating a business has now become easier and cheaper than

ever and this ties fully back into the fact that the cost of code is going to zero okay so what's the win condition here we're going to see a trend emerge where successful AI powered businesses that prioritize capturing data building usable knowledge bases optimizing their ux and then very importantly optimizing their performance with their LMS via benchmarks they are going to win okay so this is kind of a harder win condition I'm not exactly sure I'm going to measure this but but you get the general idea here right data is the most important thing then

it's the user experience right solving your domain specific problem really well for your users and then your llm benchmarks all right so one of my last predictions here we're going to see new industry break breing architecture I think we're going to see something new much like the test time compute the architecture underpinning the powerful reason models I think we're going to see something incredible like that built out by one of the big three gen companies so open Ai and Tropic or Google this is going to be a new model architecture it's going to completely change

the type of models that can be built um I do kind of see this trend toward extreme compression right everyone knows about quantization I think it might be related to that so that we can get cost and size down and make the compute Hardware more usable from a high level we're going to see a new architecture I'm keeping my eye out for this I'm looking at research papers trying to just kind of stay ahead here for a groundbreaking architecture so uh wi condition here one of the big three gench companies puts out something that significantly

improves model performance without increasing parameter count that's super super key this is one of my only low predictions here uh open AI or topic they're going to release a small language model it's going to be on device it's going to be highly performant and it's going to you know have the performance of a previous generation 2024 large language model we're going to see a scill gap earthquake I've kind of hinted at this before but we're going to continue to see the bottom of the job market Fallout that means Juniors entry level even some mid levels

it's going to be hard for entry level and Junior Engineers product Builders designers to find work at the same time going to see the top of the market rise to absurd Heights and I do want to kind of open this up a little bit and just say more generally those using generative AI effectively will experience exponential Returns on their time while those that are not are going to fall behind like very very very rapidly the wind condition here is very simple we're going to see clear job market signals showing a decline in entry level positions

while senior plus and gen AI Specialists opportunities are going to reach New Heights okay so that's 15 predictions for 2025 we'll review at the end of next year in December around the same time as this video so first off you know what is this principal AI coding is my take on the best way to learn AI coding and the whole idea here is to learn principles of AI coding to build foundational skills that allow you to excel with the a coding tools of today and tomorrow it's been 1 week since principal coding has launched it's

been really really incredible I want to start out by saying just a massive massive thank you to everyone who has used learned and is growing their AI engineering tool belt by working through principal AI coding that's what this is okay so this is the best way to learn foundational AI coding here are some of the launch stats uh we have over 300 Engineers mastering the AI coding of today and tomorrow more than 30 positive reviews it's been up for 6 days I placed the launch at the end of the video you know I just want

to shout out everyone who has you know trusted me and dove into principal a coding I put a ton of Blood Sweat and prompts into creating this and of course I could not have created it if I wasn't on the ground writing code with AI since it was first possible I've been here since the beginning you can go back on the channel and literally watch my development over the course of you know two plus years now what I've put in this course is the oil of what you need to be successful with AI coding and

of course huge thanks to everyone who's provided feedback there is going to be a price increase on January 2nd so if you're eyeing this if you're thinking about getting in just go ahead hop in there's a full refund guarantee before you start the fourth lesson and just to quickly kind of go through in a little bit more detail I built this based on a simple premise software engineering has changed it's time to change with it and we can just scroll down here AI coding is the new standard just like we were talking about today just

like in my predictions right I'm making these predictions for a reason and I am literally betting on it with my time money and energy I built it to help you avoid deprecating yourself in 2025 I do see 2024 being the last year where you can be relevant coding as an engineer without AI after 2025 it's all downhill so I built this course to help you fast track this trans I so that you can WR code with AI and be ready for the next wave which I think we're going to start seeing hint of hence my

agentic coding prediction okay in the course we focus on principles not tools everyone wants to always hype up the latest tool yesterday it was cursor today it's wind Surfer tomorrow it'll be something else same thing with models right Gemini 2.0 flash Gemini 2 experimental thinking 03 anthropic in this course you learn to endure change with principle this is a principle focused AI coding course that focuses on principles not tools we talk about the big three over and over and over this is a foundational mental model for generative AI but more specifically for AI coding if

you've already dove into principal a coding and you've completed the lessons you know exactly what I'm talking about you are already ahead of the curve there are eight lessons grouped into three sections beginner intermediate Advanced where we walk through BAS basic setup multifile editing crafting great prompts we talk about how not to suck we talk about scaling up your capabilities we talk about AER secret and then we talk about agentic loops we talk about pushing the edge of AI coding into again like the prediction I just made into a gentic coating and then we close

it off with some principled AI cating like I mentioned the price is going to go up by the time you see this it might be up it's still going to be at a great deal but it is going up you know let's hop back over to this this is here if if you want it I'm not going to keep promoting this and sharing it every single week um I just want to pitch this one more time here I'll bring it up when it's relevant to the channel on certain AI coding videos when we talk about

specific you know topics where it's relevant otherwise this is just going to be here link is going to be in the description for you if you're interested these numbers are going to continue to go up as the channel grows and you know as ad coding becomes even more relevant so why not jump in early join over 300 Engineers that are getting a massive of edge in AI coding by learning principle based AI coding that will help you stay relevant not with just today's tools and models but tomorrow's and next year's as well so that's enough

on principal AI coding there have been a couple truly just touching reviews that I've received with Engineers reaching out to me telling me that the content that I've shared over the course of two years has helped them find work build their career literally change the course of their life by you know the ideas patterns techniques and tools that we discuss on the Channel I want to say you know just massive thank you for for watching for paying attention to the channel you know not not just to the reviewers but to everyone watching right now to

you watching right now I'm here because I want to share I want to distribute this technology I want to distribute the potential and the power of this technology you know one of my favorite comments in the previous video when we launched principal AI coding was uh you know someone just said I'm buying this course based on all the value you've already provided on the channel I work really really hard every single week to make sure that I'm providing something valuable for you and for engineers so I really appreciated that comment just to get that you

know concrete feedback so without giving too much away principal AI coding is just getting started there's going to be new updates some new content some new ideas this course this resource I built for you is going to be expanding over the course of 2025 so if you are engaged with principal AI coding if you like the ideas stay tuned I'll make sure to announce it on the channel when new updates are coming to principal a coding that can push your air coding even further Beyond um thank you so much enough blabbing enough promoting this or

you know trying to sell this whatever also if you know you don't trust me yet or you know you're new to the channel and you think I'm a grifter or a sleazy sales guy I totally get it you know especially in our world with all the slop and all the you know the grifter and you know people just trying to take advantage of you I completely understand you'll learn that this is a different type of channel that what we do here is truly forward-looking um I'm going to be here every single week releasing valuable information

for you anyway let's continue let's move forward I made a a ton of predictions in 2024 I made 41 predictions and I got 36 of them correct so let's quickly just walk through some of these and review what I got right what I got wrong I'll also link last year's predictions video if you're interested in going through kind of Point by point you know full transparency I do not expect to get U anywhere above this even the 70s this time around obviously I'm hoping I do that's why I made the predictions I am betting on

them I am super convicted and I'm you know over 2025 I'm going to be putting my energy and time behind every prediction that you just saw at the same time though I realized that you know when you're trying to predict the future uh it is still uh at some level a risk a gamble so on so forth right in 2023 I got 66% right so 16 out of 24 if we look uh year over year over year okay open AI leadership I predicted that open AI continues to lead the AI race I do think that

this is still the case they are the leaders when we talk about llms when we talk about uh reshaping the industry with with reasoning models when we talk about you know everyone else playing playing copy and catchup that's exactly what we saw here over 2025 open AI released A1 and then we saw variants copies and U you know different versions of effectively what are now really well known as raing models uh open source llms I predicted that uh this is so wild to think that we didn't have GPT 3.5 level uh models at the beginning

of last year that that are open source that was true and so I predicted that that came true obviously we have models that are much better than gbt 3.5 that can run on device okay I predicted an sentience scare I marked this as wrong I don't think there were any true AI sentience scares here's an annoying one I predicted that gp5 would launch and a new architecture would launch we did get a new architecture we got new models but we did not get gbt 5 so this is an incorrect prediction so prompt engineering becomes an

essential skill for knowledge workers I think this is pretty uh non-debatable anymore the power of knowing how to prompt and the power of chat interfaces I think is Undisputed at this point uh I predicted that chat interfaces would be everywhere we'd see widespread adoption over you know massive platforms this is true every real application now basically has a some form of a chat interface okay I predicted we would see more domain specific UI for large language models we are seeing this unfortunately it's still mostly in the shap shape of the chat interface we are seeing

this starting to roll out okay so programming predictions I predicted that AI assistant would really take off uh tools like K up copilot cursor whenver AER these are now like pretty essential to a lot of Engineers workflows so you know I think 20124 you could definitely call that the year of AI assistance I think you're going to see that continue into 2025 I predicted that new programmers would be divided into two groups those learning the fundamentals and then asking questions to chat gbt and other llm chat interfaces and then those solely relying on it we

know that one of those groups has been more successful this is the way to learn you do not over rely on chat GPT you use it as a learning mechanism while focusing on the fundamentals okay top languages typescript python kind of an easy Free Bet to make honestly they're popular languages for reason they're easy to read they're easy to understand okay big Tech predictions I predicted there would be a dopamine basting Trend that kind of exploded and then Fizzles out I also predicted that Tik Tok would not get banned this was true it has not

been banned yet so I predicted that open AI would join the ranks of big tech companies I think when you think of you know big powerful tech companies now um at least you know when we do that here on the channel we are a lot more plugged into to you know generative AI in general and the generative AI companies um I think we can clearly say and clearly state that open AI is now you know a big powerful tech company with a lot of influence and it continues to increase so I predicted that Google Gemini

would briefly outpace open AI models for quite a bit of time um but this hasn't happened I I don't think even with the latest Gemini Flash and Gemini 2 experimental they are not on par with open ai's 01 model we cannot say that confidently it's getting close but we just saw open AI leap ahead again with 03 all right so I made some stock and crypto predictions here I said S&P 500 would rally and stay above 5,000 that happened I predicted Bitcoin goes above 50k uh with the surge of the Black Rock ETF that occurred

I think bitcoin's sitting around 90k right now um I predicted that ethereum would eat 20% of market dominance I got this one wrong I predicted that remote work would persist despite return attempts I think it is holding decently here we'll have to see how this trend continues over 2025 and again I kind of made a continuation bet once more this year but I do see that entry and mid level continues to be oversaturated while senior level job market job positions are in high demand uh we basically saw every frontend role disappear if you're not a

senior you know frontend specialist front-end work can be largely um automated and outsourced let's talk about maybe the most important piece of this how do we turn these bets into actions which can give us concrete results you know looking back at my my predictions this is how I'm thinking about 2025 as an engineer who wants to get maximum value for my time and effort okay so when I'm thinking about building products data is gold user experience is silver and language model benchmarks are bronze okay these are the three things that matter the most data ux

llm benchmarks I think you can safely assume everything else can be replicated by a competitor with generative AI in fractions of the time it used to take language models are not a moe the right language model plus your problem plus your data plus your ux plus your benchmarks that's your Moe get good at those things forget everything else right everything else is not defensible Junior and midlevel Engineers my advice to you as you're building up your Genera of AI skills uh you need to specialize to stand out and get hired okay a lot of people

will say go broad I don't think that's good advice I also think you should be choosing non-digital or unconventional mediums to connect with people communities companies and opportunities make sure your work is as public as possible and make it absolutely clear you are not a bot make it clear that you are a real human being right be public communicate connect let people know what you're working on use generative AI but do not rely on it for humano human communication do not automate the process of reaching out to people to connecting to uh networking don't automate

this okay part of the the reason that the bottom of the job market is falling out is because this is happening right everyone's trying to automate this that's destroying the job searching process for juniors and mid-level Engineers completely right so do not join that do not become part of the problem Reach Out try to make connections through non-digital or unconventional mediums okay so senior Engineers the shift for us is more about and and really this is for everyone okay so this is for everyone but especially senior plus push yourselves maximize your compute get in the

habit of using as much compute as possible for your work if you're coding by hand you're doing it wrong if you're coding by hand there's another engineer that is doing what you're doing much faster with compute and that means you are falling behind maybe not quickly but you are falling behind all right so what else so lock in your AI cing assistance in AI tooling but be ready to switch okay the next big thing is right around the corner AI coding assistant will become the new standard and the code editor Wars will continue and then

agentic coding will become viable right this is happening right now Devon was just released again kind of like looking at that four option pattern to how to get access to new information and tools you know the Devon Engineers are using Devon right and same with their kind of ecosystem it's now publicly available people are starting to use this technology right they have access to it we need to be ready to switch when the next big thing comes while maintaining a strong fundamental understanding of AI coding AI coding is the new standard you know just to

plug it one more time this is why I created principled AI coding this is the fast trck for Learning and understanding foundational principles to win with AI coding for today and tomorrow we focus on principles not tools right principles not tools we want to survive all of this noise all of this muck and endure change with principles okay just be ready to constantly be changing your tools and your models okay so there is no wall and again I'm saying all this from the perspective of my bets that I'm turning into actions okay there is no

wall so I'm going to assume llms will continue to improve across all dimensions and build these assumptions into my work products so on and so forth right I'm constantly spending time and energy and money on these top tier models so I can know what's possible so I can use this technology and so that I can know what's possible for the future all right and of course slop is coming so you know hold on to valuable information sources if you're not subscribed to the channel obviously I'm biased this is my channel but I highly recommend you

subscribe comment follow stay plugged in you know let the algorithm know you're interested in this especially if you made it this far into the video I've mentioned a couple of my favorite Engineers on the channel a couple times I'll do that again right now and Link them below I'm a huge fan of Simon Willison I look at pretty much every single one of his posts I check his content almost daily um he's a top tier engineer and I'm super thankful for him shout out Simon Willison and of course Paul the creator of ader all of

his work is incredible his tool is incredible uh the entire principal AI coding course would be impossible without that tool and we had to pick a great foundational AI coding tool for this course and it was a no-brainer to choose AER for this so shout out Paul those are two information sources that I stay plugged into quite laterally every single day all right lastly here constantly predict plan and act based on where the ball is where it's going and where you want to be iteratively make predictions plans and execute on your plans the world is

dynamic so your plans should be dynamic all right so let me just compress everything one more time you know if if you take nothing away from this video or if you just skip to the end for some reason it is a long video but I think it's an important video this is setting up everything we're going to be doing on the channel in 2025 here are the top takeaways here are the themes we're going to be hitting over and over again in 2025 okay so here we go max out your compute usage stay plugged in

to high value actionable information sources constantly predict plan and act ideas without action are meaningless the theme for 2023 to 2024 on the channel was the prompt is the new fundamental unit of knowledge work okay I've said this a million times I will continue to say it this is still relevant in 2025 and Beyond but we have a new theme that's popping up um that is going to be very very important the next theme and I think the big theme for 2025 max out your compute if you can gather the right context model impr prompt

and hand over your work to your AI assistants why aren't you okay okay this is a huge huge theme and if you can just focus on this one idea if you ever have the thought oh I definitely could have done that with a language model or with an AI agent or with a little bit of code or with a script that plugs into a an AI agent or working on your personal AI assistant right we're going to be talking about ADA a lot more in 2025 because the personal AI assistant is the wrapper around the

AI agent which is the rapper around the prompt right but but this is the key theme max out your compute if you can use generative AI why aren't you and if you can't use it what are the blockers remove the blockers a lot of what we're going to do on the channel is talk about removing these blockers building and using the right tooling applications AI assistants AI agents so that we can max out our compute and I mean Max it out if there is a second you don't have a prompt running it's not maxed out

okay and and and and that's how extreme we're going to be about this on the channel that there is there is a world where you have an always on machine an always on agent assistant a gentic workflow whatever it is whatever level of complexity you need imagine you have it always on always working for you building for you while you sleep this is the Northstar for the Andy Dev Dan Channel this is Northstar for all the work that I do and all the work that I do and share on this channel that's where we're headed

if you like this video Drop the like drop the sub and drop a comment let me know what you think about my predictions for 2025 thanks for watching stay focused and keep building