ɛ∞ɜɛ∞ɜ

1. We Are All Agentic Engineers Now

October 23, 2026 · 00:25:31

Read this as a blog post →

Mastered, transcribed and published by AI using these skills.

The first episode of the Millillion podcast.

Jordan Last (lastmjs) tells the story of how AI went from chatbots that made up CLI commands to agents that can generate all of the code, and why that makes all of us agentic engineers now, with an emphasis in our own area of expertise.

Transcript

[00:11] Hey everybody! This is Jordan Last, also known as lastmjs. Hey, this is the first episode of the Millillion podcast, and let me just give you a quick background on what Millillion is. I’ll give you a quick background on myself, in case you want to know that. I think it’ll be helpful. I’m a software developer. I have been a software developer for a while. I started my undergraduate education in 2009. I graduated in 2017 with a bachelor’s degree in computer science. I love software. I love technology. It fascinates me. The latent power of all these technologies that we’ve been inventing and being inspired to invent are just absolutely fantastic to me. They’re beautiful. They’re amazing. I believe technology is one of the key things, reasons, factors that brings prosperity to the world, and that helps to end physical human suffering. It’s not…

[01:22] …without its weaknesses, but I think it’s technology has been an absolute boon to mankind. Generally speaking, it’s much better than the benefits outweigh the costs, in my opinion. I’ve been professionally developing to a certain extent, probably since 2013, I would say. I’ve had a good go at it, I think. I mean, it’s not still a little bit early in my career, I would say, but I’ve had an okay amount of experience. Right now, I call myself an agentic engineer with an emphasis in software development. Before, I definitely called myself a software developer or a software engineer. Now, though, I believe personally that anyone in STEM fields or maybe even anyone in any field, we are all agentic engineers now or agentic developers. We are using AI agents…

[02:34] …to solve problems now. That’s what we’re doing. We’re coordinating agents, we’re managing agents, we’re judging the outputs of agents, and an agent, in case you don’t know, an agent is an instance of an AI, an artificial intelligence, that can actually go out and perform work for you somewhat autonomously, sometimes very very very autonomously. This is kind of the evolution of the chatbot. A lot of people, I mean since 2022-ish, AI chatbots have been pretty popular with the advent of ChatGPT. That’s generally a back-and-forth question-response type thing. I asked the AI a question in response to me, very similar to googling something in a search engine. You kind of ask a question, you have a query, and then you get a response with a bunch of results. Kind of the same thing since 2022, but late 2024 is when at least I first started to experience AI agents, and I was in the context of software engineering, where…

[03:46] …instead of just using ChatGPT or Claude to do research, and you know maybe paste a complicated Rust error into the chat interface, and you know get help, I actually was able to start to tell the AI, like in VS Code or Cursor or whatever I was using at the time, like hey I actually want you to write in here, write in my editor, do this programming test, like actually generate code, and that was crazy because it kind of worked. It worked just enough that you know I obsessed over it. This was for all, I feel like December 2024 is when I first really tried out AI agents, and you know the code, if I remember correctly it was actually pretty terrible. It wasn’t that great, but it was kind of good, and I feel like that’s like the same thing with ChatGPT, or even GPT-3, I believe in 2020 is when I first tried GPT-3, and it’s like oh my goodness, like…

[05:00] …it’s kind of good, like it seems good, and then you don’t really know how good or bad it is. I think at first you always think it’s like super super super good, right, and so you kind of get obsessed with it, and then you start going crazy with all the kinds of crazy ideas for like what you can do with this, oh my goodness, like can you generate, I mean I did a edtech company as my first little startup, and you know so I guess I probably thought like can this generate practice problems, assessments for people, and then we quickly find out back in 2020 like oh no it’s just really not that good, it’s not good enough for that. Then 2022 GPT ChatGPT comes out, I don’t know I don’t remember exactly but maybe three GPT-3.5 came out shortly after that, and then you could you know if you were using it on a regular basis or you know any basis at all you could kind of feel the jump in intelligence, right, like oh wow, like you just can feel it, like the answers feel slightly better, it’s not necessarily lying to you as…

[06:12] …much as before, but still you know in 2022, 23, 2024, if I remember correctly these things were lying to us so much, they just would say it the most ridiculous things, they would they would make up like CLI commands and API endpoints and things, they would just make things up and you’d be like oh my goodness I can’t believe this exact command that I exactly need for the situation exists, it’s so awesome, and then you go to try it and then it definitely did not exist, so that was kind of sad, but yeah anyways end of 2024 agents start actually doing things for you, like not just question response but actually generating code, and sorry train of thought, now this podcast is gonna be just whatever I want to say basically, I’m just gonna be crazy trains of thoughts and things, so you know it’s probably just going to be me for a while, but who knows maybe I’ll invite some guests on, maybe I’ll invite an AI on, and the idea is to have interesting conversations and useful conversations, I believe I have a lot of insights to share and a lot of experience to share that would be useful to many people, so…

[07:26] …yeah so this code that’s generating, it’s like it’s just good enough that maybe it could work, and at the time I was doing a lot of Rust development in the blockchain industry, building open source basically open source JavaScript and Python runtimes for a decentralized cloud basically, and so you know this is we were doing pretty I would say pretty intense relatively low level programming, not the lowest at the low level, but you know relatively low level, and you know it was actually starting to do some things, doing a lot of TypeScript as well, but it just wasn’t good enough, like I remember I spent like one or two weeks really trying to force the agent to to follow my plans, like to do what I asked it to do, I was coming up with like basically before we had the skills standard, I was trying to come up with like a skill, like just like a markdown file or something that had a process for the AI to follow, like step one do this, step two do this,…

[08:37] …step three do this, basically like you know pseudocode, trying to force it to do more than it could really do to you know I don’t I don’t really remember exactly what you know exactly what I was getting it to do, but trying to push it to become an autonomous AI agent like we have today, and it just would not follow the rules, it would just go off the rails, and it was just crazy, so I kind of gave up on that and realized like okay you know what this it’s just not ready like in early 2025 is when I was doing that it’s like it’s just not that it’s just not ready, but I could feel it I could feel it coming right like I could I could see like if things keep improving like this is gonna get really good, and even with like our taxes and accounting we were a very small team so we were you know trying to do as much ourselves as possible or at least prepare things well enough to hand off to the people we paid to do our taxes, and I remember like in in Mercury right like going through transactions, and I could feel that these agents were getting pretty good, and I had tried to you know ask tax questions before and and you would make very basic arithmetic errors, but I…

[09:50] …remember thinking like no let’s let’s prepare this I like now in each transaction like we can actually put in the notes here you know everything necessary to classify these things correctly with sales tax and use tax and like where the members were actually physically working when they did these things, and because I was preparing for a day when AI agents would be able to just grab all that information and basically do the taxes 100% themselves, and I was in early 2025 when I really felt like that that’s that seems to be coming, and then 2025 went on and into summer 2025 things were getting better agents were getting better they could do more and more but it still was really me coding and my team you know we’re coding by hand right maybe they’re the agents are generating a good amount of our code but we’re reviewing everything we’re we’re really grabbing things piece by piece and you know they started to get better they started to be able to do changes across you know more files and…

[10:55] …folders and it was getting better and better but it wasn’t until I would say October 2025 I think that’s when I really first tasted it I believe it was GPT-5.1 or maybe 5.2 and then they had like Codex versions of those things at the time as well and I was actually kind of getting out of my previous business and just applying for a job as a VP of information technology at a startup and I wanted to build like a little prototype a little demo and I realized like you know what I did then this was right around when Opus 4.5 is that the one that was crazy in October or November it was right when that crazy jump happened in the capabilities of the models and that’s when I realized you know what I think these things can generate like all of the code now like I think I can just rely on…

[12:09] …them to generate all of the code and that’s what I started doing and I started realizing holy smokes this is amazing like this we’ve jumped another threshold like now we have agents that can actually do things for us and so things got better in November things got better in December in December 2025 I actually used I believe GPT-5.2 or 5.3 to to do a very complicated tax return for the first time obviously I had a meticulous human review process like myself meticulously reviewing it I take taxes extremely seriously and it was so good it took me 24 hours though 24 hours literally because I waited till the last minute and I sat there with the AI for 24 hours building out having the day I build out crazy like verification schemes and generating all kinds of crazy code to really make sure this thing was correct and it worked at least as far as I know it worked I felt good enough to sign it…

[13:20] …it’s been accepted it seems and and we’re building on top of that so that was crazy yeah sorry I’m just just telling you my whole story now of at least AI and then I did get this job at this it’s actually a construction startup and and I basically took it on me to do a complete AI transformation at this startup and it was we’ve just been starting out so like it was basically a a fresh greenfield project right but you know in my interviews I was telling saying crazy things like no it doesn’t even matter what like yeah I have my preferences for language but it doesn’t even matter what language you write it doesn’t even matter what front-end framework you use like this AI they can just do all of it and then I joined in December 2025 and I basically started vibe coding right like just outrageous amounts of vibe coding just like I didn’t actually know like I didn’t know how really how good the code was or not because I had I had basically just been vibe coding since October because I you know it took a few months off while I was thinking what to do next…

[14:31] …after my previous company and so I wasn’t quite sure how excellent it was I wasn’t quite sure also because like the risk profile at this new at this new startup in the construction industry was very different than my previous risk profile in blockchain where I mean things have to be like as close to perfect as possible where people would lose lots large amounts of money so I was like the highest of the high standard now at this construction startup it was more like kind of a traditional SaaS apps right like building out factory software building out 3d modeling and CAD so it’s like you know what like this doesn’t have to be as crazy so I started vibe coding I mean crazy things were happening like we replaced the manufacturing software that we were going to use with our own thing I vibe coded that demo in like one day and and it and yeah so…

[15:24] …that that was like December January maybe into February but then I started to think okay like I need to start deploying this stuff into production people need to start using it and I need to protect this company from from hacks and data leaks so you know I put my blockchain hat on again and like okay this thing needs to be secure and that’s when I started realizing like all right do what do I really feel comfortable like with all this code that’s generated let me let me dig in and see no I mean I don’t feel super comfortable with it like I need to get in here and I started digging in and like you know coming up just kind of architecting the code still still from the very first day and I still believe now generating the code 100% by AI is in my opinion the right thing to do the question is after that how do you verify and in in later episodes I think I’ll go through all kinds of things about my process but back in January and February I was still reviewing code I was still architecting things you know putting in massive amounts of ESLint rules I’ll…

[16:35] …get into that process later so the point is that I have become now an agentic engineer I am an agentic engineer with an emphasis in software engineering and I was this is a point I was trying to make a long time ago in this podcast a few minutes ago that we are all agentic engineers now or agentic developers but we have an emphasis in our area of expertise my emphasis is software engineering or software development and so that is my angle that gives me the unique ability to judge the output and the verification evidence that the AI presents to me and that me and the AIs together can be super powerful and can move much faster than before and you know I personally believe we will reach AGI artificial general intelligence and I personally struggle with like at least like I don’t really believe that my software engineering emphasis is going to be vital for you know for that much longer like I don’t at some point I believe that 100%…

[17:48] …of the software engineering that I’m doing right now is going to be automated like it’s going to be so good it’s going to be so verified it’s going to have formal verification of everything it’s going to you know like I don’t know like maybe I’ll be reviewing the specifications and the behavior but I believe a lot of this is going to be super automated but for now I think my skill set is still vital and I think a lot of people’s skill sets are still vital as they’re judging right but that’s why you you need to in my opinion become an agentic engineer you need to fully embrace artificial intelligence and AI agents in a wise calculated way I think you can vibe code vibe verify vibe engineer as much as you want in the beginning and then you start to review it yourself and you start to figure out you know where the weaknesses are these these artificial intelligences are are jagged that’s a term people throw out they are jagged intelligences meaning if you think of like a graph of intelligence well what would the graph be like the y-axis is like the level of intelligence and maybe the…

[19:02] …x-axis is like I don’t maybe it’s like a bar a bar chart so it has like different areas of intelligence right like software engineering mathematics accounting art video generation audio teaching writing like there’s all kinds of areas of intelligence or at least maybe scalar capability the AI is not necessarily the same level of intelligence at all of them and there’s gaps so it makes like a jagged graph and so we as humans can sometimes fill in those gaps and that’s why we’re still useful economically for the moment I don’t believe being useful economically is like the purpose of humanity so I mean I’m not necessarily scared for the outrageous amounts of abundance that I believe we’re going to have we’ll talk about that in other episodes probably as well but yeah agentic engineering agentic development with an emphasis in your current skill set that that’s that’s that’s what we need to be doing now and for example this is kind of crazy we’re hiring a new person where I work and that person it’s the…

[20:12] …first member of my team I’ve been just a one-man software engineering team we’re gonna have two people now and this person is actually a junior developer in fact fresh just completely fresh with a degree in I believe computer science okay so what’s crazy is this person has barely any experience writing code by hand or doing any kind of software development and he learned in the AI era right so like what’s that going to be like what what am I actually going to teach him how am I going to mentor him it’s something that he and I are going to have to figure out because does does he need my full skill set I I don’t know like I mean that’s what I’m there for maybe he just needs like I mean I have a whole set of skills I have a whole pipeline to try to ensure that the slop is refined into clean code and so what’s what’s he gonna do like he could just use my pipeline and generally speaking that’s the whole point is that I’m using the pipeline obviously I’m improving the pipeline and stuff but maybe he just uses it and…

[21:23] …what he really needs to do is understand across the office the factory the construction site maybe right now he is the one to gather that context get the whole system in his head and basically just make sure the user interface and user experience is is beautiful and it works well I don’t know like what it’s it’s a new world so I but what I’m gonna do is I’m gonna train him to be an agentic engineer no matter what like he’s going to be an agentic engineer we’re going to learn how to manage agents we’re gonna learn how to write skills we’re going to learn how to you know I guess think think very critically to judge like I think judging now is one of the greatest things that we are doing as agentic engineers because basically the AIs are generating a stream of evidence to us right evidence they’re gonna generate a bunch of stuff and unless you’re vibe doing stuff like what do you want to do so sorry unless you’re generating a bunch of stuff then sorry hold on I just lost my train of thought here let me let me get right back to it so…

[22:38] …vibe engineering yes the the AIs are going to generate a bunch of of artifacts sorry okay sorry this is just a stream of consciousness so what’s going to happen is the AI is going to do a whole bunch of work for you do you really want to have to go look through that work meticulously with your own eyes in some cases you might need to but in other cases you’re going to really just want to judge like evidence that the AI presents to you for example I have a skill called basically of abstractness abstractness is a very very important property in software development and with that skill I can basically tell the AI make sure that this code like in this PR or this file whatever is sufficiently abstract and the AI will go through and you know think and reason and change the code so that it has a an appropriate level of abstraction I don’t want to have to look over all the code that it changes I instead want to trust that this skill is correct and then trust that if the AI tells me…

[23:52] …that it ran the abstractness skill then I know that that skill or that that code is abstract sufficiently that’s what you want to do and you want to keep raising the level of your own abstraction that you’re working at right like we want to give the AI ever more ambitious and bigger and bigger tasks and it will present us evidence of what it’s done it will eventually it will just say I did this correctly and we’ll trust it but we’re not there yet there’s we need to be way more in the weeds and checking but you know that’s what you do as an agentic engineer basically right you you start to you test and probe you tell the AI to do a task you look at the evidence that it presents that it did it correctly and then you judge that and then you start to gain confidence right like in the beginning you’re going to be vibing things you’re going to be maybe checking a lot of that work after but once you start to gain confidence once you have your own pipeline that turns slop into cleanliness whatever you’re applying it to then you can start to gain confidence so I think I’ll cut it here that’s the first episode of Millillion yeah I guess that’s just a little background on me and just a I want to push you to become an agentic engineer…

[25:10] …yourself in future episodes we’ll talk about what Millillion even is and we’ll dive into all kinds of concepts so thanks for listening and have a great day night or time goodbye…

Support MillillionPodcast RSS