Learning to Code

TurboTax wanted to send my tax return to India, so I built a free version of Intuit’s $5 billion software. Part 1 of a new series from Telos News

TelosTax

It’s tax filing season, and as I was making my way through TurboTax’s clunky, overpriced app, I started to get frustrated. I’d spent almost $200 on the software, and it kept interrupting my data entry with scammy upsell offers.

Then I hit a screen that blocked me from continuing until I decided whether to agree to a new set of terms. My experience using the app would be much better if I said yes. I read the tiny fine print.

Intuit, the $125 billion company that makes TurboTax, wanted all of my tax data for five years so it could share and monetize it across all of its software offerings, including Credit Karma, the fastest-growing part of Intuit’s data empire.

“Right now, we can only use your tax info to help you prepare and file your return,” an FAQ cheerily noted, “but with your consent, we can do a lot more.” I would be able to “work smarter,” “file faster,” and receive “an even higher level of service.” Buried in the last lines, I learned that my tax return data could be sent to something called the Intuit Product Development Center.

In India.

Unfortunately, the alternatives to paid tax prep software are limited. The IRS used to offer Direct File, a free online tax filing system that was popular and easy to use. But Intuit mounted a massive lobbying and influence campaign, convincing the Trump administration to kill Direct File, which makes the company’s aggressive attempts to monetize the most sensitive financial data of the 40 million Americans who use TurboTax even worse.

I found all of this extremely annoying. But over the winter, Intuit’s fellow tech industry oligarchs delivered a potential solution: AI-coding tools that enable anyone to create software.

So I decided to build my own TurboTax.

I should point out that I know nothing about coding. The last time I remember coding anything on a computer was in 1983, when my fourth-grade teacher, Mr. Bilitsky, taught our class how to use BASIC on a Commodore PET.

I have no specialized knowledge of U.S. tax law.

And yet, in just a few weeks, I created what appears to be a full-featured financial app that prepares complicated federal and state tax returns for free.

It’s TurboTax on steroids.

Naturally, I decided to call it TelosTax. The app guides you through your taxes step by step in an interview mode. If you’re a tax pro, you can enter your info directly on any of dozens of IRS tax forms. It can import your tax documents. It can organize your spending transactions by tax category. It has a dashboard where you can adjust sliders for over 20 different tax variables. It gives you a FICO-like score to assess your likelihood of being audited. The app produces pretty donut and bar charts summarizing your finances and flow visualizations that trace the trajectory of every dollar of your income, deductions, credits, and the taxes you pay.

Without any special skills, I was able to prompt coding tools to translate thousands of pages of U.S. tax law and regulations into an app that can produce a federal tax return for any American—from the simplest W-2 employee taking the standard deduction to a complex filer with, say, farm income, rental properties, asset depreciation schedules, AMT adjustments, crypto sales, a Roth conversion, net operating loss carryforward, and every imaginable combination of deductions and credits.

On the public repository where TelosTax now lives, I’m listed as the “author” of the app’s 234,000 lines of code, despite the fact that I did not write even one of those lines. I asked my new robot friend, Claude Code, the AI that wrote the whole thing, how long it would have taken a human to do the job.

“Without AI assistance,” Claude said, “I’d estimate 2-3 years for one person with the right skills.”

At first, that flattering response made me marvel that I had built something this complicated in just a few weeks.

But then it filled me with dread, because it hinted at the obvious limitations of using these tools without the proper training in coding or tax law. Two-hundred-thirty-four-thousand lines is a lot of code. Like good writing, good code is concise. The frontier AI models—Claude, GPT, Gemini—are trained to be useful and compliant, which makes them sycophantic. Did Claude actually build me a bloated, unmaintainable pastiche of features that proves the opposite of what I intended—that the big software companies have nothing to fear from vibe coding amateurs?

If the app holds up and works as advertised by Claude, I’ll never need to use TurboTax again. But the reverse could be true: maybe Claude, in its effort to please, created a Potemkin app that simply fooled me into thinking it was in the same league as a product designed by hundreds of engineers at a Fortune 500 company.

To help answer that question, I’m making the app public as a free open-source project so tax professionals and actual human coders can vet it. The idea behind this series is to explore two related issues:

• Whether new AI-coding tools are really up to the task of replacing premium software

• How journalists can use AI, which everyone seems to hate, for a positive public-interest goal

So consider TelosTax an experimental prototype and don’t use it to prepare your taxes unless your return is carefully reviewed by a professional. It’s possible Claude got everything wrong. If you’re a tax professional or programmer who would like to help stress-test the app, please email me at ryan@telos.news. But anyone can play around with it, dig into the code, and contribute to the project. You can check out the code and documentation on GitHub, and the app is now live at its new home: https://telostax.app. I’ll report back on how it holds up, and whether Intuit has anything to worry about.

The rest of this series is divided into four parts that track my own circuitous journey exploring these issues over the last month:

• Part 1 (Today): How Intuit Killed Direct File — The story of Intuit’s attack on a popular program—and how the Trump administration gave the company everything it wanted

• Part 2 (Wed): Why Everyone Hates AI — To build my own bespoke software, I had to confront my knee-jerk hostility to AI, which I had come to associate purely with slop and an impending jobs apocalypse—starting with journalism. I was wrong (mostly).

• Part 3 (Thu): AI Comes for the Coders — A look at how coders have navigated the end of coding and the lessons that journalists—and every other white-collar profession—can learn from them

• Part 4 (Fri): The AI Rabbit Hole — My own weird experience falling down the rabbit hole of AI-coding to build a tax app

It’s been a fun ride, and I’m glad to finally get to share all of the details with you.

---

Part 1: How Intuit Killed Direct File

Intuit CEO Sasan Goodarzi. In December 2024, Intuit donated $1 million to Trump’s Inaugural Committee. A few months later, Trump killed Direct File, the government tax-filing program that threatened Intuit’s $5 billion in annual revenue from TurboTax. (Getty Images)

Why we can’t have nice things (tax-filing edition)

Intuit’s invasive privacy policies sent me on an odyssey to understand how the hell they get away with this. The simple answers are: money and Trump.

Since 2015, Intuit has spent $33 million on an influence campaign in Washington, D.C., to make it more expensive for Americans to file their taxes. The company’s number one goal, as laid out in lobbying disclosures and SEC filings, was to stop the federal government from providing a free, simple tax-filing system.