Back to all articlesDMS JOURNAL / INSIGHTS
Agentic Era7 min

Six AI Uses From One Week’s Experiments

Installed AI but Stopped There? Six Practical Uses From a Week of Experiments

This retrospective review of eight full-length videos and seven Shorts published from the morning of October 9 through the morning of October 16 summarizes six AI experiments, their limitations, and the process of checking results.

Six AI Uses From One Week’s Experiments
DMS / VISUAL ESSAY

This is a republication of Reedo’s Naver article. It is not a new experiment or a DMS client case study. The cover and in-article photographs are AI-generated staged scenes, not evidence of actual work.

If you have installed AI tools but still have no idea what to ask them to do, looking at finished results is a quicker starting point than learning new model names. This video covers adding alarms to notes and turning a text-only homepage into a 3D space you enter by scrolling. That does not mean it all happens at the press of a button.

I reviewed the eight full-length videos and seven Shorts published on Today’s Me | Reedo AI Log from the morning of October 9 through the morning of October 16. Grouping Shorts that covered the same exercises with their full-length videos, I put together six uses worth trying at the time.

AI-generated staged image, not an actual work screen.

How were search and team operations delegated to AI?

The starting point is to assign specific things to check rather than simply ask for answers. The Aside episode follows seven steps, from installation to connecting a model, permissions, memory and skills, scheduling, connecting coding tools, and sending instructions from a phone. When making a request, specify the target, the desired result, and when to stop. It is also important not to immediately treat the numbers in a retrieved table as verified facts.

Episodes 2–5 of the dots series walk through recording a brand’s tone and restrictions in a brand folder, having a skill read past posts, and then splitting the strategist and writer roles. The final checks are whether the files exist, whether the restrictions were followed, and whether the text was actually read. This is an exercise using a fictional brand, not a case that demonstrates real sales or growth results.

Aside from installation to instructions by phone

dots Episode 2 Brand folder

dots Episode 3 Skills and past posts

dots Episode 4 Two AI employees

dots Episode 5 Delegation and three checks

How far did the photo editing and 3D homepage get?

AI-generated staged scene 2AI-generated staged scene 2View original

AI-generated staged image, not an actual work screen.

The photo-editing exercise used LightCraft and PhotoCraft to work on 10 Canon RAW photos, import existing Lightroom presets, and export the results. It also revealed limitations, including dark bands along edges when removing haze and some preset information not carrying over. Rather than calling it a complete replacement, these are reasons to preserve the originals and start by testing one or two copies.

The 3D homepage work went through Claude Code’s plan mode, a rules file, and Blender, leading to a page where the camera moves as you scroll. More important than an attractive opening screen was actually scrolling through it to check whether each section lined up with the camera position. The house and architecture studio in the video are fictional settings for practice.

Editing 10 RAW photos and testing presets

From a text-only site to a 3D homepage

Was an entire app really copied?

AI-generated staged scene 3AI-generated staged scene 3View original

AI-generated staged image, not an actual work screen.

The key to this sticky-note alarm experiment was studying how the app worked and building something new, rather than copying it wholesale. REA was connected to Claude Code with conditions that actual notes must not be opened and the original source code must not be copied. Function snippets appearing during analysis did not mean that a complete app had been replicated as-is. An experiment for personal learning must be distinguished from copying and distributing someone else’s software without permission.

The process of adding a note alarm that was not there before

Are longer prompts better for AI videos?

In Flow, three prompts were each tested three times to recreate the same bottle. In this small test, a short sentence explicitly specifying a cylindrical bottle preserved the shape in all three attempts. But the wording differed as well as the length, so this does not mean shorter prompts are always better. The information about free credits also reflected the conditions at the time of filming. It is better to check the current screen for credit costs by resolution and duration, and begin with a small test.

Comparing prompts to recreate the same bottle

What to check before using Flow’s free credits

You do not need to start all six at once. Choose one task you repeated today, write down the result you want and what should not be done, and compare the original with the result side by side. The common thread that week was not a promise that AI would do everything, but a process of narrowing what to delegate and checking the results again.

Original Naver article

Reedo on YouTube

Reedo on Threads

Reedo portrait

Reedo Insights

Translating technology into practical language

With over 19 years in 3D design, optical communications equipment development, and global field training, I now connect AI automation, creative imaging, and practical channel operations to document ways of making complex work simpler.

Newsletter

New writing,
in your inbox.

Receive notes on AI, automation, and building income. The newsletter is currently sent in Korean; English articles are available here on the blog.

New articles only · Unsubscribe anytime

Start a conversation

Turn an idea into something practical.

Whether it is automation, design, training, or content, we can start with the problem you need to solve.

Get in touch