GPT-6 Astra: 15 Use Cases to Learn from Global Power Users
When AI-driven residents were placed in a game, they started talking to each other.
When a photo of a house was provided, it was transformed into a 3D home complete with furniture.
Meanwhile, it is also being used for professional tasks like triaging customer inquiries to the right representatives and handling initial video edits.
What makes looking at global GPT-6 Astra use cases so interesting is this sheer breadth of application.
It doesn't just end with "getting a good answer"; what you create remains on the screen. The AI is operating the software you use every day.
In this article, I will introduce 15 ways to use it based on public posts, detailed reviews, and demonstration videos.
We will look at everything from creating houses, towns, and games to refining web pages, handling inquiries, editing videos, and drafting articles.
By the time you finish reading, you will have decided on at least one task you want to delegate to Astra.
I'll take you that far.
First, Understand the Difference in "Tools" Given to the AI
The official ChatGPT account has guided GPT-6 Astra's capabilities in PC operation, web browsing, development, science, and professional work.
https://x.com/ChatGPT/status/2095597504226267333
However, the cases introduced here cannot all be reproduced by simply typing a single sentence into a standard chat box.
Global users are letting the AI use tools and materials they have on hand, such as 3D modeling software, video editing software, and customer management services.
Since this includes reports of early access, the features and pricing available may not be the same as what you see on your screen right now.
This article focuses on "what was provided and what was created," without listing unconfirmed pricing.
We are counting 15 use cases, including different applications by the same individuals.
I have separated the results reported by the posters from the suggested applications for use in Japan.
01 | AI Game Residents Starting Conversations with Each Other
First is an experiment by Matt Shumer.
He requested the creation of a world using Unreal Engine, a game development software, and used Astra to drive the people within it.
The people collaborated to survive in that world. The goal was to create residents who not only talk but also act within that space.
https://x.com/mattshumer_/status/2095596175705399482
The next day, Matt posted that he heard human voices and thought someone was in the room. The source of the voices was the residents of the world he had created.
Hearing this alone might make it seem like the AI spontaneously developed consciousness.
Reading the detailed review reveals that the conversation between residents was part of the originally requested behavior. This isn't a story about the "birth of unexpected life."
Still, it is striking that he conveyed the conditions of the world he wanted to build and got multiple characters to move and converse.
Furthermore, it wasn't smooth from the start. There were issues like animals falling into the water, and the world was nurtured by correcting behaviors.
There were also overlapping conversations, so it remains an experiment with some rough edges.
Another important point is preparation. Matt separated the Codex managing the progress from the Codex handling the implementation and used existing character assets. The conditions are different from just typing one sentence into a blank chat.
If you're trying this in Japan, start with a "conversation between two people" rather than a whole town.
For example, for customer service practice, place a customer and a clerk and test the response when a return request comes in. For storytelling, move two people who have different opinions on the same event.
The first thing to decide isn't a long description of the characters' personalities, but rather "what the conversation is meant to decide."
If you have it output the final agreed-upon content, it turns from a fun experiment to watch into a form usable for practice or drafting.
02 | Turning a House Photo into a 3D Home with Furniture
Tom Krcha introduced a way to turn a photo into a 3D house.
He reported creating a house in Blender, a 3D production software, and placing furniture, appliances, and even toys.
https://x.com/tomkrcha/status/2095598645190291775
It didn't just output a single finished image.
It was created as a 3D shape that humans can adjust later, and he explained that it can be moved like a game on his own terminal.
This is the difference from simply generating an image of a room.
"Move the sofa a little to the right," "Make the desk smaller." There is room to proceed with such changes by adjusting the 3D parts.
A familiar application in Japan would be furniture layout before moving.
Provide a photo of an empty room and the dimensions of the desk or shelf you want to place. Have it create a plan for placing the desk by the window and another by the wall.
Comparing the two side-by-side is easier to communicate to family than just comparing them in your head.
For stores, this can be applied to plans for changing the positions of product shelves and cash registers. It serves as a draft for comparing how things look before actually moving heavy shelves.
However, the AI doesn't know what's behind the walls not shown in the photo or the exact dimensions.
The more realistic it looks, the easier it is to forget that.
If you're using it to actually buy furniture, measure and provide the room width, ceiling height, and doorway positions. For parts where dimensions are unknown, leave them so they are clearly marked as temporary values.
When requesting, instead of "reproduce the photo perfectly," it's better to say, "make the known parts 3D and list the unknown dimensions," so you know what to check next.
03 | Walking Through a House That Hasn't Been Built Yet
If the previous example was "making a house 3D," this one is "walking inside that house."
Chris introduced a demo showing a scene where a house created in Blender is brought into Unreal Engine 5, allowing you to walk through the house before it's built.
https://x.com/ChrisGPT/status/2095594374843232583
Converting 3D house data into a form that can be explored from a human perspective. That is the process introduced in this demo.
Since I cannot confirm that Chris himself created the entire process, I will treat this as an "introduced demo."
Thinking about a meeting using this reveals its utility.
You can not only say, "this hallway feels a bit narrow," but also imagine entering from the front door and going to the kitchen with groceries.
Even if the furniture layout fits in a floor plan viewed from directly above, it might feel cramped from a human perspective.
If applying this, a use case for checking a single movement is easy to understand.
Walking from the entrance to the desk. Moving from the store entrance to select a product and going to the register.
By creating just that short range, what you want to see becomes clear.
Tell the AI not just to "make a house I can walk through freely," but "I want to check from a human-height perspective from the entrance to the destination."
Specifying the furniture to place along the way and the locations you want to check reduces the gap where the AI builds a gorgeous exterior but misses the areas you actually need to see.
Of course, the fact that a walkable video was created is different from confirming the safety of the architecture.
If you can show the areas of concern in a video, it becomes easier to consult with architectural experts. It's material to communicate "I want to pass through here like this."
04 | Building a Manhattan Streetscape Over One Week
Matt Shumer again. This time the stage is the streets of Manhattan.
He has released a video showing how he used Astra to build street by street within Unreal Engine. According to his post, the production spanned one week.
https://x.com/mattshumer_/status/2095609734845927525
The interesting thing is that he didn't output a large image of the city all at once.
The detailed review explains that he first built up a street and then expanded from there.
Here, too, a system was used that separated the side managing the progress from the side implementing it. He also provided direction along the way, such as prompting it to be more bold.
In other words, it doesn't mean "a perfect Manhattan was completed in one week without touching it at all."
Production is ongoing, and the video includes fast-forwarded parts. It's a case where the short video in the post and the actual work time spent should be viewed separately.
Still, there is potential in the method of gradually building a large object like a city.
If we were to apply this, a corner of a shopping street or a single aisle of an event venue is enough to start.
If you want to compare the positions of stalls, you don't need the whole city. You can start the conversation if you can see the entrance, the aisle, a few shops, and places where people might gather.
In the first section, decide on the size of the buildings, colors, and density of signs.
If you like it, use those criteria for the next section. After expanding, check the joints and places that are impassable.
This order can also be applied to article sites or internal documents.
By deciding on one page as a finished sample first and then increasing them, you can minimize the scope of realizing "the atmosphere is different from what I thought" after everything is done.
Rather than imitating large-scale production as is, there seems to be more use in taking home the "agree small and then expand" part.
05 | Creating a 3D Fish Game Playable in a Browser
Theo has released a 3D game that can be played in a browser.
The video shows a space like an aquarium with 3D fish swimming around. Water plants and rocks are also placed.
https://x.com/theo/status/2095599934766764338
He explains that it was created with a single instruction.
However, the post alone doesn't reveal the full text of the instruction or the materials provided before it. It's best not to assume that "my single sentence will definitely result in the same thing."
What we want to take away here isn't the story of making a game big.
Create something you can understand by interacting with it a little, instead of reading an explanation.
For example, an observation screen for small fish for aquarium lovers. When you select a fish, its name and habitat appear.
For school learning, make it a quiz where you place creatures in the correct locations. For product introductions, replace it with an experience where the finished image changes when you choose a color.
These are application ideas from the article side, but if you narrow down the purpose to one, the required screens will also be fewer.
You don't need to add scores, member registration, rankings, and billing from the start.
"Open," "Select one," "Response returns." First, complete this short flow.
Instead of asking the AI to "make an interesting game," it's better to say, "make a one-screen prototype where an explanation appears when you click a fish," so you can recognize when it's finished.
Before going public, keep it in a form that only runs on your own computer. You can decide whether to distribute it to others after trying it out.
Because there are more things you can make, deciding on features not to add at the beginning makes it easier to finish the prototype to the end.
06 | Creating a Minecraft-style Block Game as a Prototype
Next is a Minecraft-style game introduced by Flavio Adamo.
The post describes it as a "Minecraft one-shot." The video shows a block world with roads, fields, and trees, along with tools in hand.
https://x.com/flavioAd/status/2095597137849446688
What we want to distinguish here is "being able to prototype a familiar play style" versus "completing the same thing as a commercial game."
You can't say that everything from saving, communication, and stable operation over long periods is included from a short demonstration.
On the other hand, the use case of bringing the prototype of a play style you want to create into a form you can actually touch is easy to understand.
Rather than writing "you can build buildings freely" in a planning document, having a single screen where you can place blocks communicates what kind of game it is to enjoy.
If trying this in Japan, instead of reproducing a famous game, change it slightly to suit your own purpose.
For example, a game where you place desks and shelves as square parts to create your ideal work room. For children, a screen where they try to build a bridge with limited building blocks.
The first completion conditions of "place," "delete," and "return to start" are sufficient.
Rather than making the appearance close to the original, checking if the selected operations can be done without hesitation will lead to the next improvement.
In the request, write the description of the screen along with the operations you will test.
"Place one part, delete it, and confirm that you can redo it." If you ask this much, you'll know what to look for after being told it's finished.
Prototyping based on a famous work and selling it using that name or materials are also different things.
At the stage of considering publication, replace them with your own name, your own rules, and materials that are allowed to be used. Don't mix learning the mechanics of play with bringing the work over as is.
07 | Turning Historical Battles into 3D Scenes You Can Watch in Motion
Dan Shipper of Every introduced the Battle of Waterloo as a subject.
He evaluates it as being created with a single instruction and having historical reproducibility.
However, the historical accuracy is the poster's evaluation. The results of cross-referencing the unit placement and movements at each time with other historical materials have not been confirmed.
What's interesting about this case is that it turns history from "something to read" into "something where positional relationships are visible."
Things that are hard to understand in text, like who came from where and what distance they moved, can be shown as a space.
The same idea can be used for explaining difficult procedures.
For example, in a manufacturing site, the order in which materials enter, are processed, inspected, and shipped. For an event, the order in which visitors move from reception to their seats.
When it's hard to grasp positions from a text-based procedure manual, replace it with simple 3D shapes and arrows.
You don't need to make the people look realistic; it's useful if you can see "wait here" or "carry to here" even with boxes or circles.
In that case, what you should provide first is not the wish for a "cool video," but materials with the correct order written down.
Also, tell it not to arbitrarily supplement actions not in the materials and to display unconfirmable parts as temporary.
Moving explanations can be persuasive even if they are wrong. It's safe to include a check for just the order and position separately from the visual check.
Not just 3D for viewing the finished product, but "3D for explaining to someone." This use case has a place even for people not interested in games.
08 | Building a Walkable City by Combining Prepared Buildings and Roads
You don't need to create houses, roads, and trees all from scratch.
In another official demo introduced by Chris, Unity, a game production software, is used to build a city by combining existing materials.
https://x.com/ChrisGPT/status/2095601996770263362
The video shows a city with blue buildings and palm trees. You can see the progress between buildings from a ground-level perspective.
This method of assembly can also be confirmed on OpenAI's official page. Chris is introducing that official demo.
This is slightly different from the previous "making a house 3D from a photo."
Assembling into the place you want to create using the parts you have on hand. What you ask the AI for isn't just the generation of the materials themselves.
If applying this, a layout plan using event equipment is a familiar example.
Desks, panels, chairs, product shelves. If you have all the materials you're allowed to use, provide them and create a small exhibition space.
Enter from the entrance, see the products first, and hear the explanation in the back. Have them lined up according to that order.
With just "a futuristic and cool venue," you might end up with equipment you don't actually have.
Provide what can be changed and what would be problematic, such as "use only these parts," "don't block the entrance," and "be able to walk to the reception."
First, open one plan and see if products are hidden or if there are obstacles in the aisle. If it's wrong, change the layout while keeping the same parts.
This is an application idea from the article side. It doesn't mean the safety check of the venue or the permission to use the materials is complete.
Before drawing everything, try lining up what you have now. For that purpose, you can narrow down the scope of the request and the place for judgment more than creating a single finished city.
09 | Creating a Game Where You Walk with a Purpose, Not Just Look at the Scenery
Peter Gostev released an adventure game where you walk freely through a wide area.
The post includes the description "GPT-6-Astra (Ultra)" and an open-world adventure game.
https://x.com/petergostev/status/2095596341422440714
The video is built from the player's own perspective. There is dry ground and trees, a list of objectives is displayed on the left of the screen, and the name of the destination is shown at the movement site.
You can also see a camera in hand. However, features that aren't clear from the short video, such as whether the captured images can be saved, cannot be determined.
The time it took to create or the full text of the instructions are also unknown from this post.
Still, there are places to pay attention to when watching. In addition to the beautiful scenery, "where you have come" and "what you should do" appear on the screen.
Even if you create a walkable world, if the person who enters doesn't know what to do, they will just move a little and be done.
If trying this small in Japan, you can make it a guide game that visits three locations.
For example, walk through a fictional store and see the product shelves, workspace, and pickup counter in order. A short explanation appears at the place you arrive.
If you know you've "seen everything" at the end, it's easier to grasp how far you've progressed than a simple 3D tour.
A guide game visiting three points is an application idea from the article side based on how the screen is shown.
When requesting, decide on the "initial objective," "display upon arrival," and "how to end" before increasing the terrain.
Instead of "make a vast adventure game," it's better to say, "make a prototype that visits three locations and ends when all are seen," so you can confirm it yourself to the end.
If you're going to have people touch it, also check if they can return to the entrance if they get lost. The things you need to add for the people playing aren't just the number of buildings.
10 | Continuing to Build a City-Building Game Over 5 Days
Matthew Berman reported that he used Astra early and tested games, text, and PC operations.
Among them is a city-building game reminiscent of SimCity.
https://x.com/MatthewBerman/status/2095595893991129444
The video shows houses, tall buildings, and cars running on the roads. Around them are columns for viewing the state of the city and buttons for selecting operations.
The highlight is that he's creating the entire screen for playing, not just the city scenery.
He explains that he used "/goal," which gives a goal and continues the work, and ran it for 5 days.
Furthermore, even at the time the demo was released, the work was reportedly not yet finished.
It's best not to read this as "a finished product on par with a commercial game was completed in 5 days." How many times a human intervened in the middle is also unknown from this post.
On the other hand, it's an interesting example as a report of continuing to build a prototype of some scale even after outputting the first screen.
If we were to try it, instead of running it for 5 days, break it down into forms you can touch along the way.
First, to the point where you can place one building on a vacant lot. Next, to the point where you can delete the placed building and redo it.
Furthermore, decide on a budget and make it so the remaining balance changes every time you place a building.
This order is a proposal from the article side, but if you open and check them one by one, you can notice states where only the appearance has progressed and the play isn't established.
In the request, also write "first give me a version where I can place and delete one building." If a working version remains, you'll know what was accomplished even if further development stops.
When running for a long time, confirmation of usage and costs is also necessary. Don't apply the features and execution conditions of early users to yourself as is.
Being able to request a large game and whether you should request it large from the start are different things. You can decide whether to continue while touching the intermediate deliverables.
11 | Providing a Sample Screen and Fixing Page Text and Margins
"The moving page is done. But it's somehow hard to see."
This is the stage where you want to use the screen creation introduced by OpenAI's official account for developers.
Provide a rough sketch, a screen you want to use as a reference, and the current page, and create a moving screen based on them. It's a way to adjust text size, placement, margins, colors, and operations.
https://x.com/OpenAIDevs/status/2095596149654868092
The official explanation also mentions showing the screen in the middle of work as an image and using it as a clue for corrections.
The basis for this chapter is the official use case explanation. Below is an example of applying that request method to a course information page.
If applying this to your own page, it's easier to request if you don't end with just one word like "make it stylish."
Provide the current screen and a sample you think is easy to read side-by-side.
For example, if it's your own course information page, communicate the places you're having trouble with, such as "the headings are small," "the price is hard to find," or "the text is cramped on a smartphone."
The sample is material for consulting on how to handle colors and margins. It's not for bringing over other companies' text or logos as they are.
You can also narrow down the first correction to just the top of the page.
"Don't change the product name and description, but fix the headings, margins, and the position of the application button." This way, you can compare what changed before and after.
After correcting, open it in both PC and smartphone widths. Confirm that text isn't overlapping, the price can be read, and buttons can be pressed.
If you have the corrected parts and confirmed operations returned, the places you should review yourself will also be decided.
Of course, fixing the appearance alone doesn't guarantee that sales will increase. First, it's a way to communicate hard-to-read places with words and screens and confirm the corrected results yourself.
12 | Triaging Inquiries to Representatives and Even Creating Reply Drafts
From here on are cases closer to daily work.
Claire Vo is having Astra operate a customer management service screen and reconfiguring the flow of inquiry response.
Divide who will respond according to the company size and the content of the consultation. Create a reply draft that also includes a link for scheduling, in that person's name.
Originally, it was a job of connecting parts on the screen to create a process. Claire is delegating the parts that take time for a human to operate, including the browser itself.
In the video, she handles items for creating emails, such as the recipient, subject, and body, in addition to triaging the representative.
What's important here is that the AI is not arbitrarily sending them to customers.
The reply draft is passed to Slack, and a human confirms and sends it. Automatic sending is mentioned as a future stage.
Even if you don't delegate all the way to sending from the start, you can reduce the effort of reading, triaging, and drafting.
Applying this in Japan, you could consider a use case of dividing inquiries to a course into "price," "schedule," and "enrollment conditions" and preparing response drafts.
To do that, what you provide are good past reply examples and content that is okay to answer. You don't need to let the AI decide on unknown prices or special handling.
For example, decide "don't propose discounts not in the materials" and "divide consultations that cannot be judged into 'needs confirmation'."
Even this alone makes it clear what to focus on when reading the reply draft.
The first practice is enough with a few inquiry texts with personal information removed. Decide whether to connect it to actual business after seeing if the triage and drafting are correct.
The good thing about this case is that the list of pending responses returns in a "state where a human can judge," rather than a flashy app being created.
13 | Finding Problems to Fix First from Scattered Customer Voices
This is an example from Claire's video that is easy to think about applying to work.
Inquiries, meeting notes, development records. She is creating a function to summarize what is happening with a product from information in different places.
She is handling information from Intercom, Granola, Linear, GitHub, documents, and more.
You don't need to remember the names. It's enough to think of it as a state where information is scattered across tools used for customer response, meetings, and development.
From there, it reads the content, organizes duplicates of the same story, and summarizes them in a form that can return to the evidence.
The issues appearing in the video are also specific.
The AI stops in the middle. The function for users to cancel by themselves isn't working well.
There are also requests to integrate with other business tools.
With this, you can specifically decide on the next job, like "let's fix the problem where it stops first."
Regarding this function, Claire explains that most of it was done with the first instruction, and she added further instructions and screen adjustments afterward.
It's not a perfect one-shot completion. It's an example of finishing it by putting in manual work along the way to a state where she can use it herself.
If trying this small in Japan, you can start with just this month's inquiry list.
Divide questions by similar content and list the troubles that are appearing repeatedly. Attach the original inquiry number to each item.
Furthermore, have it separate "problems with many cases" and "problems with few cases but requiring urgent response."
For example, if there are 10 voices saying the guide is hard to understand, while there is 1 voice saying it can't be used after payment, it's hard to decide by simple case count alone.
This is an example for the article, but the meaning of providing what you want to prioritize becomes visible.
If there is a list that can return to the original voices, you can also confirm when you have doubts about the AI's summary. It's a way to decide the priority yourself while looking at the collected evidence.
14 | Delegating the "First Rough Edit" of a Long Video
Dan Shipper reported that he had Astra operate video editing software to perform the first edit of a video handling Fable 5.1.
The rough edit mentioned here is the stage of lining up materials and first making it into a form where you can see the whole thing.
What he states is only up to the first edit; it's not a story about delegating everything from subtitles, volume, finishing, and publishing without correction.
Specifically which cuts were chosen and what kind of instructions were given are also unknown from the post body alone. Think about the use case without filling that in with imagination.
For people who make videos, the task of first looking at all the materials and searching for the parts to use is a big burden.
Before the sense of the finished video, the road to getting to a state where you can start editing is long.
If applying this, start from the "organization of parts you want to show" in a course or dialogue.
Provide the original video and transcript, and have it find the locations for "explanation of price," "demonstration," and "answers to questions."
List the start time, end time, and reason for keeping each. Once the parts to use are decided, rough edit them in that order.
If you make this into one request, a human can enter from the point of confirming the candidates. You can separate the task of searching for places to use from the task of deciding on adoption.
However, if you only pull out a single exciting word, the conditions before it might be dropped.
If you delete the part that was explaining "only in this case," the meaning of the story changes.
Separately from whether it could be made short, confirm if the meaning matches the original statement.
This is why you add "keep not just the conclusion, but also the conditions under which that conclusion holds" when requesting.
Try it with materials you filmed or materials you're allowed to edit, and don't change the original video. If you keep it as another editing plan, you can go back even if you don't like it.
It's a way to reduce the time spent starting from an editing screen where nothing is lined up, while deciding the final presentation yourself.
15 | Creating the First Draft of an Article and Starting from Where the Writer Corrects
Finally, text. However, to end with "it can also write text," there was a specific report.
In Every's Astra review, it is clearly stated in the beginning, which can be read for free, that the first manuscript was written by GPT-6 Astra from a single instruction.
The article's byline is Katie Parrott and GPT. Dan did not write that manuscript himself.
In that beginning, an exchange is also posted where Dan thought it was a text written by Katie.
For the reader, it looked like the usual writer's text. That point is the interesting part of this case.
However, it doesn't mean the AI knew that person's life.
The materials used for input and the corrections added by a human are unknown from public information alone. This article also does not speculate on the content of the paid portion or non-public instructions.
If applying this yourself, what you provide first is not "write like a pro," but materials that actually existed.
Things you tried, things that didn't go well.
Questions that came from readers, the method you finally chose.
And, attach your past articles as a reference for sentence length and endings.
The point is to separate them as "references for style, not the facts for this time," without letting it reuse experiences written in past articles for the new article.
Once the draft returns, see if there are increased events you don't know about or impressions you didn't say.
The cleaner the text, the more you want to skip that confirmation. So, ask in advance: "Don't write experiences not in the materials, and let me know the missing parts."
The value of the article is the materials and judgment the writer had. Delegate to Astra the parts of putting them in an order that can be read and creating the first draft.
Reduce the time spent worrying in front of a blank page and start from correcting whether what you want to say is correct. In writing, this use case is familiar.
Looking at 15 Examples, Where Will You Start?
The world where residents talk and Manhattan are eye-catching.
But a person who wants to make tomorrow's work easier doesn't need to create the same scale from the start.
The criteria for choosing are having materials on hand and being able to tell for yourself if it was done.
If you have photos, a room layout plan. If you have inquiries, triage and reply drafts.
If you have videos, candidates for scenes to keep. If you have notes, the first draft of an article.
None of them require thinking of a new big project before using AI. You can use one task that is currently stalled.
If it's hard to choose, please check the following three:
- Are the original photos, documents, videos, or data on hand?
- Can you confirm the quality with your own eyes?
- If it's wrong, can you return to the original state?
Tasks where all three are "yes" are easy to think of a first way to try.
Conversely, tasks where you don't even know if the answer is correct and are sent outside as is are things you don't have to choose first.
Note that flashy demonstrations also include the work of preparing tools that can be used and the process of a human correcting in the middle.
If learning from global cases, look at not just the finished video, but also the materials provided before it and the parts a human is looking at at the end as a set. That way, you can replace it with your own work.
Initial Request Text You Can Use by Just Rewriting
The following is not a reprint of the prompts released by the people introduced.
It is a template organized for this article to request your own work. If there are features that cannot be used in the current environment, I have it set to let you know first.
What I want to do:
[What to make, what task to reduce]
Materials to provide:
[Files, URLs, samples that are okay to use]
Please distinguish between the style and appearance of the sample and the facts to be used this time.
Scope to complete first:
[A small range, such as one room, one screen, 5 inquiries, etc.]
Please confirm the tools that can be used and let me know first if there are operations that cannot be done.
Conditions for judging as finished:
[Specific operations you will confirm or required output]
How to proceed:
Think of the necessary tasks and save the finished product to a separate file.
Do not create missing information; distinguish between unknown points and temporary values.
Confirm according to the conditions and fix problems that can be fixed.
Things not to delegate:
Do not overwrite or delete original files, send externally, publish, or purchase.
If additional costs or new access rights are required, stop before executing.
Things to return at the end:
The location of the finished product, what was confirmed, and what cannot be confirmed yet.
Please summarize briefly the points a human should look at next.
For example, if it's inquiry organization, fill it in like this.
What I want to do is read inquiries to the course and prepare for replies.
What I provide are 5 inquiries with personal information removed, the course guide, and past reply examples.
The first completion scope is just those 5 cases. Divide into price, schedule, enrollment conditions, and others, and create a reply draft for each case.
The conditions for completion are that the original inquiry number remains, prices not in the guide are not created, and cases requiring judgment are separately identifiable.
What I have returned are the triage list and reply drafts, and I won't let it send to customers.
With this, what to confirm when opening the finished product is decided. It's a request that's easier to judge whether it's usable than "do something convenient."
First, to the Point of Receiving One Finished Product
These 15 examples are not meant to make using AI difficult.
Provide a photo and make a room. Provide an inquiry and make a reply draft.
Provide a video and make candidates for editing.
Provide a note and make the first draft of an article.
The entrance is surprisingly clear.
Choose one first and provide the materials on hand. Complete it small and open it to confirm.
You only need to expand to the next task for the things that were usable there.
Choose one unfinished task you have now and fill in the blanks in the request text above.
That's the first thing to do after closing this article.
*This article is composed based on public posts, personal reviews, and demonstration videos as of September 4, 2026.
The results introduced are reports from the posters and do not mean that everything was reproduced in this environment.*
