AI codes better than me. Now what?

It's 2026 and it feels weird to say, but I'm not really writing code by hand anymore. I mean, I built this entire

anymore. I mean, I built this entire Rustbased image compressor that compiles to web assembly and uses a speltkit web app to drop in images, compress and

resize them. I built all of this without

resize them. I built all of this without writing any code by hand, entirely using coding agents. I even built this game, a

coding agents. I even built this game, a Space Invaders clone, where I didn't know anything about how the hardware worked. I was able to work with coding

worked. I was able to work with coding agents to write Verarilog code and actually program the hardware on the device which is just crazy. I mean six months ago or a year ago I would have

never even tried these projects and now they were actually very easy for me to do. So I want to talk about 2026 what

do. So I want to talk about 2026 what this means for the future of software and kind of where I've been in the last half a year. So about nine months ago, I posted a video on this channel about how

I tried out a bunch of text editors and idees and I ended up picking Cursor to use because I felt it had the best UX. I

was familiar with VS Code, but also was doing a really good job on the AI features. A few months later, I decided

features. A few months later, I decided to actually join Cursor. So I'd been there for about 6 months now, primarily because I could really feel the landscape changing. I could feel

landscape changing. I could feel software engineering changing from underneath me and I wanted to be part of figuring out what that thing was going to be and help teach you all how to

navigate that journey yourselves. So, I

haven't really posted anything for the last 6 months or so. I think there was only one video and a lot has happened.

Just a year ago, coding agents weren't really a thing. Cursor's agent had just came out and then in 2025 we saw Cloud Code and Codex and so many other coding agents get very popular. coupled with

the models themselves improving. So

going back a year ago, pre-Sonnet 4, the models hallucinated a lot, they were pretty bad at following instructions and calling tools. And when you look at the

calling tools. And when you look at the latest crop of models, uh the Codeex Maxes, the Opus 4.5s, which I'm sure will continue to only improve, they've got really good. I mean, it's pretty

remarkable how well they can generate code now. And it has me reflecting a

code now. And it has me reflecting a little bit on what this means for building software in general. I mean, if I'm being honest, Sonet 4 was probably better at coding than me than I want to admit. So, the models have probably been

admit. So, the models have probably been good enough for a while, but you couldn't use them reliably well for most of the software engineering practice.

So, now going into 2026, I would say the current crop of models are definitely better at coding than me for sure, and they're allowing me to take on more ambitious tasks. But this has kind of

ambitious tasks. But this has kind of changed for me. What is software engineering again? Like, writing code

engineering again? Like, writing code was never really the bottleneck. uh

especially for larger projects. So if

writing code is basically free now or it's basically easy to do, what does that mean for the rest of software? But

before I get to some of my recommendations, I want to show some of the things I built in the past few months to help contextualize why I feel like software is changing so much. So

the first thing is that I actually moved cursor.com off of a CMS to just markdown and MDX files. And I thought it was going to take a long time. Actually, I

thought it might take weeks. we might

hire some contractors or an agency and I was able to get it done pretty quickly, you know, in a matter of days and for a very affordable amount of money relative to the impact that it had, you know,

within the Cursor Ultra plan, which is pretty wild. That was my first

pretty wild. That was my first experience where I thought, "Wow, I can probably be much more ambitious with what I asked coding agents for." So then the next project I talked about, my

rustbased image compressor, again, I did this entirely with coding agents. I did

not write any code by hand. I thought

for sure the models would not be good at writing Rust, so lower level code versus like TypeScript and also doing more complicated computer science things, figuring out these algorithms, writing them from scratch. I mean, if you go

look at the source code, it's doing some pretty wild stuff. But then the last one is also this game I've been working on.

So I have the Analog Pocket, which is a kind of like a Game Boy, and you can actually write code and program the hardware that's on the device. This is

something I have no experience with. I'm

not a game dev. I built some games in Python, but that's it's very different.

Many layers of abstraction hire. And I

was able to ask the model and the agents questions and get significantly far with this to where I could build new games from scratch now now that I understand how it works. So my level of ambition

has continued to rise in what I can build. And when I think about 2026, that

build. And when I think about 2026, that trend is only going to continue. So,

here's what I would recommend for for you all watching and as you're thinking about, you know, your job as a software engineer or maybe you're writing code for fun, how to think about the next year of progress. First, I think the

natural reaction to all of this is, yeah, well, I work on this one thing where the model isn't very good at it yet, so that means that the model will never get good. And I kind of felt that way for things like Rust or lower level

code or, you know, detailed computer science topics. And then recently, the

science topics. And then recently, the models have just got really good at them. So I if I was you personally, I

them. So I if I was you personally, I would expect that within the next year the models will get good at most things that you need to write code for in software engineering. Now that doesn't

software engineering. Now that doesn't mean that software engineering is dead or that we're not going to have jobs or anything like that. I actually think it's the opposite in some ways. I think

there will be more demand than ever for software engineers and people to actually use these tools. And it's a little bit of a paradox, but I think it's played out in history a bit. So

what you need to understand is that in learning these tools, whatever coding agent you want, cursor or codeex or cloud code or whatever, in learning these tools, you're going to get a significant advantage over those who are

maybe more hesitant or skeptical or they're not really sure if they want to use these tools. And learning how to use them well can put you significantly ahead. Now, obviously I'm biased because

ahead. Now, obviously I'm biased because I work at Cursor. You don't have to use cursor. You can use any of them. But I

cursor. You can use any of them. But I

think the entire profession of software engineering is changing and it's worth really earnestly giving these tools a chance and trying to figure out the best way to take advantage of coding agents.

So the the second thing to think about here is you want to use these tools to get rid of all the drudge work, the stuff that you don't want to do. And

when you think about it, there is a lot of software engineering that is effectively like copying and moving JSON files or figuring out why a shell script didn't work or trying to debug some strange error message. Like there's a

lot of boilerplate. There's a lot of things you just copy paste and you previously put in Stack Overflow or Google. Use coding agents for that

Google. Use coding agents for that stuff. Like you can get rid of all of

stuff. Like you can get rid of all of the boring stuff and focus on the things that actually matter for the end product. Now that code is cheap, what

product. Now that code is cheap, what really matters is your taste and increasingly as an engineer that is blending over to be more of a generalist. So it's not only code is

generalist. So it's not only code is what is the product that we're building?

How does it look? How does it function?

What's the UX? uh how are we going to sell this thing? How how do we market this thing? And a great way for you, my

this thing? And a great way for you, my kind of third tip here is to go build things, be a generalist and actually try the entire gamut of building something.

Don't just write the code and publish it, but actually get it out into the world, market it, get users, build something that people love and grow a product, figure out the right things to build. Those skills are going to be very

build. Those skills are going to be very valuable for you as a software engineer for the next one to two to three years.

As this profession is changing, the product engineers who embrace AI to do their job better are going to have a significant advantage. I have a whole

significant advantage. I have a whole post on product engineers if you want to learn more about how I think about this, which is kind of funny because the trend was already happening back when I wrote this in 2024, but yeah, it's drastically

expedited now with AI going into 2026.

The fourth and final thing is to take time to really learn. I think if you rely on LLMs too much, they can kind of trick you into thinking that you understand a topic and you don't want to

delegate your thinking to the agents, to the models. You want to use them to

the models. You want to use them to become smarter. So, how do you do that?

become smarter. So, how do you do that?

I think it's helpful to ask the models lots of questions. It's like you have this great pair programmer that's sitting beside you all the time that you can ask unlimited questions to that's never going to get mad at you or think

that it's a dumb question and you should use that to your advantage. One thing

I've been doing recently that's really fun for example on my two latest projects is I'll ask the model to generate basically a mini course for me to generate guides walking me through

these topics that are completely new to me. Explain to me what Verilog is and

me. Explain to me what Verilog is and how the syntax works. explain to me the different layers of abstraction from the hardware to the actual code that I'm writing and how it gets compiled. Like

these things can make basically ondemand learning much more accessible versus having to go read like 10 textbooks to figure a lot of this stuff out. So if

you have the agency and the will to go learn these things, man, you can get just a huge advantage right now. So

highly recommend spending the time to actually go and learn these things. And

if you want to learn even more, you know, shameless self-promotion, but I do have some things here that might help.

First off, uh, leer robb.com/ai. If you

want to learn more about how LLMs work under the hood, very concise posts.

Hopefully, that'll be a little helpful.

And then also, if you want to learn a bit more about the foundations of AI as applied to coding models, cursor.com/learn, I have a little course

cursor.com/learn, I have a little course here as well that walks you through things like what are tokens, what's a context window, and I want to make more stuff here. This is kind of it's kind of

stuff here. This is kind of it's kind of my job now is I want to help engineers make this transition from the pre-AI world or the the midAI world into the future where coding agents can write

most software for us and what does that mean for the profession of software engineering. So if there are things that

engineering. So if there are things that you want to see me teach or talk about, I would love to hear it. And honestly,

I'm figuring this out just like you all are. It's kind of weird a little bit of

are. It's kind of weird a little bit of like sometimes an existential crisis because it feels like you're losing a part of the thing that you really love which is building things. But for me, I

had to remember it wasn't about the code. Writing code was never the

code. Writing code was never the bottleneck in a lot of ways. It's about

building something great and something that I'm proud of and something that I think is really good. And it's so much more than just the code for that. So

now, how can I use AI to be even better at my craft that I have this great tool?

So yeah, that's what's on my mind for 2026. It's going to be a great year. I'm

2026. It's going to be a great year. I'm

optimistic. I think we're going to have lots of positive things. But let me know what you think and what you want to see me teach about. Peace.