In July 2026, I wrote The Rising AI Sea.
Today, I want to share my first trivial use of a “bot”.
More precisely, Grok Bot, which comes free with an X Premium+ subscription.
Grok Bot is a persistent agent that can use a computer and connected services. In the cloud.
Grok Bot is like a super-smart robot helper that lives on the internet.
“Persistent” means it doesn’t forget you. Regular chatbots forget everything when you close the window. This one stays awake and can keep working on your task even if you walk away or turn off your computer.
“Agent” means it’s not just talking — it can actually do things. It has its own computer (that lives in the cloud, which just means “on the internet, not on your laptop”) and it can use that computer to open websites, click buttons, move files, and follow instructions.
“Connected services” means it can plug into the apps and websites you already use — like your photo album, email, or calendar — once you give it permission. Then it can clean up your photos, check your email, or do other jobs for you while you’re busy.
So basically: it’s a robot helper that lives online, remembers what you told it, and can actually use a computer to get stuff done for you.
This example is trivial, because it is a bot to run few tasks. I have approximately 90,000 photos in Amazon Photos and I want to remove exact duplicates first.
Key requirements:
- Identify exact file duplicates (same content/hash).
- Prefer to keep the copy that is in an album (if one is in an album and the other isn't).
- Delete/move to Trash the duplicate that is not in any album.
- Do not permanently delete yet; move to Trash so it can be recovered if needed.
- Generate a clear review report of proposed deletions before actually moving anything.
- Work carefully with a large library; batch if necessary.
- User is in Canada (amazon.ca likely).
I asked Grok to outline its plan for how it will access the library (browser login, unofficial API with cookies, etc.), how it will detect album membership, and the steps for the first scan. It must ask for any needed credentials or confirmation to proceed with logging in. It will NOT delete anything until I have reviewed and approved.
It took a few minutes for the bot to be generated as: Amazon Photos Cleaner.
After the exchange of credentials between the bot and Amazon Photos, the Amazon Photos Cleaner started working and sent an Excel file for my review:
I reviewed the potential removal of 1000s of duplicates for the year 2025.
I approved the duplicate removal.
The Amazon Photos Cleaner went to complete its task: Trashing the images one at a time in my Amazon Photos account. I turned my computer off, receiving confirmation of the tasks completed on my phone.
Is this “programming”? Coding?
Not in the traditional sense. I didn’t write a script, choose a language, or debug loops and edge cases. I described what I wanted in ordinary language: find duplicate photos in my Amazon Photos library for 2025, show me the candidates, and remove the ones I approve.
The bot turned that request into a working process. It produced an Excel file I could review, waited for my
decision, and then trashed the approved images one by one. My role was closer to supervising a capable assistant than writing code: define the goal, inspect the intermediate result, and give the final go-ahead.
That distinction matters. Traditional programming is mostly about implementing the steps. Here the implementation was handled for me; the valuable part was the judgment — deciding which
near-duplicates were safe to remove and which ones I still wanted to keep. It is the same pattern I wrote about earlier: AI is good at volume and repetition, while a person still supplies the context and the final call.
So, is it coding? Only if we stretch the word. It is closer to directing. The skill is shifting from knowing the syntax to knowing how to state the outcome clearly and how to check the work before it runs. That feels like a smaller, more accessible form of programming — one that does not require a computer-science background, just a clear idea of what “done” looks like and the willingness to stay in the loop.
Your mileage may vary. But for a first, low-stakes experiment, it was enough to clean thousands of photos without me writing a single line of code.
Good botting!