How to Clean Up a Messy Downloads Folder Using Codex

I hate messy things. To me, happiness is an empty mailbox. But there is one place on my MacBook Pro where my discipline has clearly failed: the Downloads folder.

My wife has a Google email account with more than 10,000 unread emails. I would guess that most of them are commercial messages trying to sell her something. I am very different. I have to confess that I have seven email addresses, but my inboxes are almost always clean. I rarely leave anything unprocessed.

The Downloads folder, however, is not an inbox. It is more like a digital junk drawer. On my MacBook Pro, I had more than 3,000 files and folders sitting there. Some were useful. Some were old. Some were duplicates. Some were mysterious. So I decided to use Codex as my cleanup assistant.

Step 1: Count the Mess

The first thing I asked Codex was simple: How many files and folders are in my Downloads folder?

Codex can run shell commands locally, so this is exactly the kind of task it handles well. Instead of manually opening Finder, scrolling forever, and guessing, I could ask a plain English question and get a concrete answer. In my case, the count was roughly 3,000 items.

Step 2: Measure the Space

Next, I asked Codex how much disk space the Downloads folder occupied. The answer was about 100 GB.

That is a lot of space. Fortunately, my disk is 1 TB, so I could tolerate it. But tolerance is not the same as organization. One hundred gigabytes of random downloads still feels like a warning sign.

Step 3: Find a Photo I Barely Remembered

This was the coolest part of the experiment. I remembered that I had taken a picture with Dr. Ming Wang and a lady, but I did not remember the filename. I did not remember the date. I only remembered the scene: one lady and two gentlemen.

So I asked Codex to help me find a picture with one lady and two gentlemen. It diligently searched through the images and returned exactly what I wanted. That felt different from ordinary file search. I was not searching by filename. I was searching by memory.

This is where AI becomes practical. A messy Downloads folder is not just a storage problem. It is also a memory problem. Codex can bridge the gap between what I remember and what the computer actually stores.

Step 4: Build a Small Cleanup App

After seeing the scale of the mess, I asked Codex to write a Python program with a graphical user interface. I wanted a small tool that would let me review files and decide what to delete.

This is one of the most underrated uses of AI coding tools. You do not always need a polished commercial app. Sometimes you need a small personal utility that solves one annoying problem in your own life. Codex can write that kind of software quickly: list files, show metadata, preview candidates, and help me delete only what I decide to delete.

Step 5: Find Duplicate Files

I also asked Codex to find duplicates. Sometimes I download the same file more than once. A duplicate detector can compare files by computing a cryptographic hash, such as SHA-256. If two files have the same hash, they are almost certainly identical.

This approach is much better than comparing filenames. Two duplicate files may have different names, and two files with similar names may not be the same. Hashing looks at the actual contents of the file. In my cleanup, Codex found duplicates and helped me save about 6 GB.

Why More Developers Are Paying Attention to OpenAI

This small Downloads-folder experiment also reflects a bigger trend I am seeing among developers. More people are comparing AI coding tools carefully, and many are moving some of their work from Anthropic's Claude Code to OpenAI's Codex experience.

The first reason is practical: developers care about availability, speed, and how much work an agent can do before it hits limits. OpenAI has enormous computing resources, and many developers feel that translates into fewer interruptions, less downtime, and more room for long coding sessions.

The second reason is model quality. Model comparisons are always partly subjective, but many developers believe GPT-5.5 is simply a better coding partner than Claude Opus 4.7 for the kinds of tasks they do every day: reading a codebase, planning changes, editing files, running tests, and staying inside the workflow.

The Bigger Lesson

Cleaning a Downloads folder may sound ordinary, but it shows something important. AI agents are not only for building big applications. They are also useful for small, personal, annoying problems: counting files, measuring disk usage, finding old photos, detecting duplicates, and writing tiny utilities that make your computer easier to live with.

The lesson is also bigger than this one folder. AI is already beyond the chatbot age. Agentic AI is moving at full speed. A few months ago, I tried Roo Code, Google's Antigravity, OpenClaw, and Hermes Agent. From my personal experience, Codex is the best. Mark the date: May 8, 2026. In one or two years, the situation may be different. But I am telling you the truth at this particular moment.