What Happens When You Give Coding Agents $100 to Build a PDF Editor

Building functional software using modern artificial intelligence models has become a popular experiment among developers. Recently, one developer tested whether four autonomous coding agents could design and build a useful PDF editor from scratch when given a small financial budget. The results offer a fascinating look at how machines approach software creation compared to human developers.

While the resulting applications compiled successfully, they often suffered from frustrating usability quirks. Understanding these shortcomings highlights why human oversight remains essential in software development. It also shows why specialized utility tools often beat out generic artificial intelligence creations for everyday tasks.

The Experiment and Its Setup

The developer tasked four coding agents with creating an open source PDF editor. Each model received a modest budget of one hundred dollars and explicit instructions to inspect existing tools. They were told to search online for common user complaints and build something genuinely intuitive. The models operated within private environments to ensure they did not influence one another during the process.

Most models ultimately decided to construct web applications rather than native desktop programs. This choice immediately introduced accessibility hurdles. While web tools can run anywhere, they often lack deep system integration. Users expect document tools to open instantly and handle local files smoothly without relying on constant internet connectivity or complex browser permissions.

Furthermore, the models struggled to anticipate how everyday people actually interact with documents. Human users are purpose driven and impatient. We expect tools to respond instantly and place elements exactly where our cursors point. Artificial intelligence agents, by contrast, evaluate static screenshots rather than experiencing real time friction. This fundamental difference in interaction leads directly to awkward interface decisions.

Interface Quirks and Navigation Flaws

When artificial intelligence designs user interfaces, the results can be delightfully bizarre. In one instance, an agent attempted to make its design exceptionally clear by placing a heavy textual description on every single button. Instead of feeling intuitive, the layout felt cluttered and overwhelming. It ignored established cultural design languages that humans recognize instantly.

Other models went in the opposite direction, packing the screen with countless tiny icons that left users guessing. None of the models thought to include a search bar for settings or utilities. Modern operating systems rely heavily on search functions to help users locate tools quickly, yet this modern convenience was entirely overlooked by the autonomous builders.

Performance and Feature Limitations

Performance degradation was another common issue across the generated applications. Without a human pushing the limits of the software during early tests, backend requests often bogged down the user interface. Simple tasks took seconds to process rather than happening instantaneously. When dealing with multipage documents or heavy graphics, these delays become intolerable very quickly.

Feature sets were similarly restrained. Most models built basic viewers that handled simple annotations or page reordering, but struggled with advanced manipulations. Editing pre existing text or managing complex vector graphics proved largely out of reach. When features were included, they often came with frustrating bugs. For example, some applications inverted colors in dark mode or skewed images upon export.

Practical Document Management Steps

When you need reliable results without fighting buggy artificial intelligence experiments, relying on established browser utilities is often the best approach. You can manage your files securely without uploading anything to external servers. Here is how you can handle common document tasks directly in your browser:

1. Visit the PDF Converter to transform your images into clean document pages or merge multiple files together.
2. Open the PDF Compressor if your file is too large to email, choosing a gentle clean up that keeps your text crisp.
3. Use the PDF Splitter to extract specific page ranges or break a massive document into smaller sections.
4. Fix any upside down scans instantly by heading over to the PDF Rotator to adjust your pages by ninety or one hundred eighty degrees.

Why Human Oversight Still Matters

The gap between what artificial intelligence can code and what humans actually enjoy using remains wide. Coding agents excel at writing boilerplate logic, but they lack taste, patience, and empathy for the end user. They do not sigh when a button requires three extra clicks, and they do not feel the annoyance of a jagged scrolling motion on a mobile screen.

Until artificial intelligence systems learn to experience software the way humans do, we will continue to see quirky applications that require a steady human hand to polish. For everyday creators and small business owners, sticking with privacy focused, purpose built browser tools ensures your workflow stays smooth and frustration free.

Frequently asked questions

Why did the coding agents build web apps instead of native software?

The models likely chose web applications to ensure cross platform compatibility. However, this choice made it harder for the tools to integrate natively with local operating systems and file managers.

Did the AI generated PDF editors work well on mobile devices?

Most of the resulting applications were broken or poorly optimized for mobile platforms. Issues included jagged touch scrolling and display ratios that cut off interface controls entirely.

What major features were missing from the AI built editors?

Common omissions included settings search bars, optical character recognition, advanced text editing for pre existing content, and bug reporting mechanisms.

How can I edit PDF files privately without using buggy software?

You can use free, secure in browser tools that process everything locally on your device without uploading your sensitive documents to any external servers.

Sources

1. LLMs still cannot detect bad UX - nielstron (blog.nielstron.de)
2. PDF - Wikipedia (en.wikipedia.org)
3. List of PDF software - Wikipedia (en.wikipedia.org)

How we made this post: we researched the public sources listed above and drafted the article with the help of an AI writing tool (Google Gemini). Automatic checks look for copied wording, numbers that are not in the sources and broken links, but they cannot catch every mistake. Spotted a mistake? Tell us.