AI Helped Me Write Over A Quarter Million Lines of Code. The Internet Has No Idea What’s About to Happen.
Austin Starks · Originally published
8 min readIf you ask 10 developers what they think of ChatGPT and AI-first code editors, you’ll get 10 different answers.
For one, we have the old farts that are resistant to change. These grown-ass men cry all day about how ChatGPT is garbage and glorified autocomplete. In reality, these people are just anxious that the skills they developed over a lifetime just became worthless overnight.
Then, we have the junior developer still in college. These people scream about how ChatGPT is the best thing since sliced bread. Meanwhile, when you take away their cheating IDE, I swear they can’t even update the dependencies in a useEffect array.
As someone who has written literally hundreds of thousands of lines of code, both with and without the help of my “glorified autocomplete” here’s my take on how this technology will transform software engineering.
Permanently.
Proof that I am indeed a coding prodigy
Now, I know with the title, the bitter Redditor is WAITING to jump out at me and say
well ackhsually, there’s no way one person wrote that much code.
They are calling me a liar from the safety (and anonymity) of their keyboard.
But as someone with a MS from the literal best school in the entire world (Carnegie Mellon University) IN software engineering, I HAVE absolutely written this much code.
And what’s better?
There’s proof.
austin-starks - Overview
A man with many interests, including deep learning, computational medicine, and algorithmic trading. Cornell and CMU…
github.com
I have released a half-dozen AI tools to the community. Most of them are public and have to do with AI, Finance, and the intersection between the two. So if you want to see how much code I’ve written, you can just… look?
This is also perfectly visible with my GitHub contribution history.
In the past 1 year alone, I’ve had exactly 3,798 contributions. That’s more than 10 per day on average.
I am a coding fanatic.
But the actual proof is cold, hard numbers. So let me dispel your disbelief once and for all.
I used the open-source cloc tool to count the lines in the selected source directory of my app, NexusTrade.
NexusTrade - No-Code Automated Trading and Research
Perform financial research and deploy algorithmic trading strategies
nexustrade.io
NexusTrade is an AI platform that helps retail investors automate their trading decisions and perform comprehensive financial analysis. Using the cloc tool, we can count how many lines of code are in each directory. I’m going to use it on my src directory, which:
- EXPLICITLY excludes
node_modulesand other dependencies. Whether a configuration file is counted depends on its file type and the command’s exclusions. - Breaks it down by language — TypeScript, Rust, JSON, CSV, and more.
And now, for the moment of truth.
The cloc output reports 263,646 lines across the selected files, including more than 133,000 TypeScript lines and over 20,000 Rust lines. The included language list also contains JSON and CSV. Those can hold configuration or data, so this total is a repository-size measure rather than a count of executable code or lines independently authored by me. The screenshot does not supply a complete split of generated files, data and configuration.
Note that this is just one project. The reality is I have worked for years as a software engineer and most of my code is in my old job’s repo. I also have a variety of different open AND closed-source projects that runs to tens of thousands of lines of code. Take, for instance, my platform to help me manage my LLM prompts. This is 28,000 lines alone.
That one-project count sits alongside my other public, private and employment work. It shows the scale of the systems I work on. My claim about how AI changed the workflow comes from the process described next; the count alone does not measure time saved or AI’s share of authorship.
The next edgy keyboard warrior is getting ready to type out the following. I literally hear their fat fingers slamming down on their dirty, crumby Mac.
If you wrote that much code, your project sucks! It’s not that complicated
I swear these internet developers have never created a complicated system in their life. Once they see that my app has an LLM-interface, they think it’s a 500 LOC toy project that they can make in a weekend.
To them, I’ll say the following.
Do it.
Replicate my app in a weekend. Implement all of the features. Hell, make it open-source and free. Put me out of business!
I dare you.
Because this app is extremely complicated. The people making these comments can’t even wrap their head around a fraction of the platform. The app has:
- 17 different worker jobs that run at a regular cadence to do some work, whether that’s process incoming orders, send marketing campaign emails, hydrate fundamental data, or nudge users to complete tutorials
- 40+ different routers and controllers to respond to requests from the client
- A 20,000 LOC custom-built highly efficient algorithmic trading engine written entirely in Rust. Have you ever coded in Rust before? It’s like getting boxed every time I open my computer
- Over 33 different pages implemented in React. I don’t even feel like counting the number of custom, re-usable components these pages use. It has to be several dozen.
I spent 18 months rebuilding my algorithmic trading platform in Rust. I’m filled with regret.
medium.com
Algorithmic trading is complex. Did you expect the app to fit in a single JS file?
Naturally, if you’re an intellectual developer, you’re probably wondering… how the hell do you write hundreds of thousands of lines of code by yourself? Especially with the help of AI?
Here’s how I actually pull it off.
What is my coding workflow?
AI is my junior engineer slave that I get to boss around and pay pennies to for complex tasks.
(And don’t give me shit. I can make that joke because I’m black)
There is a catch though – the engineer is more like a bootcamp grad that learned from reading books and doing Hello World applications. Thus, you have to dictate it around and hold his virtual hands.
For smaller tasks, it actually does fairly well on its own. I use Cursor’s agentic features to write the code for me, and then I always give it a good read with my two human eyeballs, as well as dropping the git diff into a good model.
Larger tasks are similar but much more involved. I usually have multiple tabs open; one tab for Claude 3.7 Sonnet and another for Gemini 2.5 Pro or ChatGPT. Honestly, with the world moving so fast, it changes every month.
I’ll then ask the AI models the question. I’ll do this by:
- Copy/pasting ALL relevant code snippets
- Giving the model similar examples so it understands what the architecture looks like (for example, it knows I’m using TypeScript with a standard MVC design patterns and a MERN stack)
- Breaking down the task into EXACTLY what I want
- If applicable, give relevant articles or notes about what the task is about
With this system, I have the power of an Indian freelancing agency (with the upside that the only communication barrier is between me and a half-sentient, schizophrenic ocean-boiling machine). While I was HIGHLY productive without LLM tools, these tools have completely transformed how much work that I am capable of doing.
Migrations that would’ve taken weeks of agonizing labor literally took days now. Landing pages that would take HOURS and would come out looking like a Freshman former-CS info sci major made it 19 minutes before the Canvas deadline locked them out.
I’m not a frontend engineer. What can I say?
Instead, I get polished, high-quality beautiful code that just works. Even writing an entire article comparing 5 different LLM models took me a few hours. Then, I integrated the final result into my platform.
This article shows a real-world example of what I mean 👇🏾
Despite all of the hype, Google BEATS OpenAI and remains the best AI company in the world.
O3, O4-mini and 4.1 are not enough to dethrone Google
medium.com
The fact that my little engineering buddy can code all day with no breaks (and no benefits plan) is insane. Whether you like it or not, it WILL transform software engineering entirely.
What is the future of AI with coding?
Now, just because the job role will change doesn’t mean they’ll go away. The people that say otherwise need Lexapro; I mean, did compilers eliminate the job of a programmer?
AI just makes it possible for us to get shit done. We can do the Corporate America version of a 3 month task in 3 and a half hours. But that doesn’t mean all code that we could ever write will be written. That literally doesn’t make sense.
For example, once I’m rich and famous from my NexusTrade platform, I’m going to make a video game company. It will be a MMORPG that’s the combination of Runescape, Skyrim, and Avatar the Last Airbender.
And I’ll make it with AI.
Back in the day, if you wanted to make a video game company, it would’ve taken you decades and cost millions. From the voice actors to the writers to the devs and everything in between. A four-person shop couldn’t even kinda pretend to release games.
(That’s a lie, actually. I’ll change that to “fix games people want to play”.)
Software is unlimited. And thanks to AI, anybody can bring their crazy ideas to life. And that’s truly wild.
And if you’re reading this article thinking “thank God I’m not a software engineer, I’m completely safe from AI!”…
You would be DEAD wrong.
AI is not just coming for engineers. It’s coming for investors too!
AI is going to transform every single industry, from marketing to sales and engineering. But most importantly, it’s going to absolutely transform how smart people make their investing decisions.
Instead of paying millions to a financial advisor (who has never outperformed the S&P500 at his 20 years at Northwestern Mutual), smart investors are now using AI to help make smarter investing decisions.
In fact, some of the strategies created by AI are literally unbelievable. This one, for example, is the historical backtest linked in this April 2025 article. The original teaser claimed “600%” relative outperformance, but this excerpt omits the strategy return, benchmark return and exact dates needed to reproduce that ratio. Open the saved portfolio to inspect the record; its later state is separate from the original snapshot.
I asked Google’s Gemini 2.5 Pro to create a trading strategy. It earned 30% in the past year.
As someone who went to the #1 AI School in the world (CMU), this result is jaw-dropping
medium.datadriveninvestor.com
Imagine a more intelligent agent that can update its strategies with the market day-by-day. This future “isn’t coming”. It’s here, right now.
The time is NOW to learn how AI can help you make smarter, data-driven investing decisions. Try out NexusTrade for free and see the difference AI really makes.
NexusTrade - No-Code Automated Trading and Research
Perform financial research and deploy algorithmic trading strategies
nexustrade.io
Or sit on the sidelines and do nothing! I’d much rather make money, but the choice is up to 😂!
No comments yet.