GPT-6 Astra + Work: esta combinación es muy bestia

0:00This is GPT6 Astra, the most powerful artificial intelligence model to date, especially good at working on complex tasks. And this is ChatGPT Work, the mode in ChatGPT that lets you assign complete jobs without needing advanced technical knowledge. When you combine the two, the result is truly beastly. You can do more things, work faster, and achieve more complete results. And that is precisely what we are going to put to the test. So in this video, I'm going to show you how to start using GPT6 within ChatGPT Work and some of the most interesting uses for this combination. One of the most interesting ways to use Astra is

0:41precisely within Work. If you don't know it yet, you can find it inside ChatGPT. You select Work and you'll be inside this mode. If you are used to using the usual chat, this will seem quite familiar to you. You still have a box where you write what you need, but inside Work, the good thing is that you can assign complete jobs from start to finish. For example, imagine you have to prepare a proposal for a client. In the chat, you can write, "Give me ideas on what to include in this proposal," and from there, you work on the ideas in the conversation. In Work, the task

1:14can be: here are my client notes, prepare the proposal, and save it in a document. You are asking it to work with that material and prepare the file for you. And then you can tell it, " Change this part or add another option, " so it keeps working on it. There are things you can do in both modes, but for the examples in this video, where we are going to assign complete jobs, we will use Work. For the demonstrations, I will use the desktop

1:39application, in my case from Windows. If you don't have it yet, I'll leave the download link in the description. Install it and log in with your ChatGPT account. Once inside, verify that you have Work selected. Here below the box where we write, we open the model selector and choose GPT6 Astra if it is available in our account. We can also choose the reasoning level. This controls how much time and effort the model dedicates to solving what we have asked of it. A higher level can help it prepare more complete responses and solve more difficult jobs. With a lower level, it usually responds faster and consumes less usage. Because increasing

2:16the reasoning also has a cost, and it's not necessary to set it to the maximum from the start. As OpenAI's head of core products explains in this tweet, GPT-6 Astra with low reasoning effort already outperforms GPT-5.6 with high reasoning effort. If you had good results with high reasoning, their recommendation is to start with Astra on low or medium. So, to begin, we'll leave it on medium, and we can always

2:40increase it later for a specific task.

2:47One of the most interesting things we can do with Astra is analyze many files at once. Imagine a folder where you have documents for your business, projects, notes, and even logs of meals or workouts. We can give it access to study all that information as a whole. For this example, we are going to focus on a business. I have prepared a folder for a company called Lumbre Studio, which makes videos and photographs for other businesses. Here we have the company information. In this folder are the invoices and payment records. In clients, we have agreements and conversations, and in operations are the deliveries, time logs, and team

3:23notes. We're going to ask it to cross-reference all those documents and tell us what's working well, where the problems are, what activities are taking up our time, and what we should prioritize. There are several ways to give it access. We are going to do it by creating a local project. We check that we are working locally. We click

3:41on choose project and create a new one. We'll give it a name, in this case, Lumbre Studio, and then select the computer folder we want to use. In this case, we search for the company folder, select it, click open, and finish creating the project. Now we have it here in the projects section. We can return to it whenever we want and open a task within it. The tasks we open in this project can work with the linked folder, and here next to the message box, we can check which project we have selected. With that access, it can read the documents and also save new files or make changes when we ask it to. For

4:18this first task, we are going to ask it to analyze the information and create a report while keeping the original documents. This is the prompt I am going to use. I ask it to review all the files, find problems and strengths, and prepare an action plan for the next 30 days, the next 3 months, and the next year. We want it to review finances, clients, and work organization, and to back up its conclusions with the documents; if it finds a problem, it should indicate where it comes from and what information proves it. We also ask it to compile the entire report into a PDF and save it within the project so that

4:52we can open and review it. We'll send it and see what it returns. Alright, after almost 5 minutes, we have the report. And if you haven't worked with this mode yet, there are a couple of things I want to show you. What it's done is create a file with the report inside the project. If we go back to the folder we connected, here are the original documents, and inside the results, we have the new report. This is what I was saying; it can read the files you work with and create new ones from them, but you don't need to open it from the explorer. Here in the chat,

5:27we click on the filename, and we can view it within the application. Let's enlarge it to read it better. And as we can see, it has prepared a nine-page report for us. At the beginning, we have a summary with priorities, the supporting information for each, and the next steps. Then it details the issues and the plan to resolve them. When preparing this company profile, I left some problems scattered among the files to see if it would relate them, and it found them here. For example, we have a supplier invoice for € 726, but the transactions show two full payments with the same reference. The report links both records and detects

6:06that € 726 was overpaid. The next step would be to verify this with the supplier and request a refund or agree on compensation. We also have a client, Ruta Cero, with two overdue invoices totaling € 3,630 including VAT, and it verified that the corresponding work was already delivered and accepted. In the case of Órbita Café, there are four extra videos that the client approved and received, but they don't appear as invoiced. That’s € 700 before VAT, € 847 in total. Notice that to find this, it had to cross-reference the client agreement, the conversation where they accepted the extras, the delivery logs, and the invoices. The information was spread

6:47across several documents. It also found an organizational issue. Lucía, one of the partners, spends 19 of her 40 weekly hours on revisions and administration. And there is a project that sat waiting for review for several days, so it suggests scheduling specific times for reviews and defining which changes the editor can approve directly. The report also covers what is working well, such as accepted deliveries and repeat clients, and ends with a plan for the next 30 days, the next 3 months, and the next year, indicating who should handle each action, by when, and how to track progress. That’s the idea: leverage the documents you already have to find

7:26issues you might be missing and have a concrete plan to start resolving them.

7:35Another thing we can use Astra for in Work is writing, for example, to turn a complicated document into an explanation that someone unfamiliar with the topic can understand. But if we want it to always follow our criteria, we can create a skill. A skill, if you aren't familiar with them yet, is basically a set of reusable instructions. This way, you don't have to explain how you want it to work every single time. You can find them here in the side panel, under add-ons and then in skills. This is where the ones you have installed appear. And to create a new one, we just have to open

8:08a new task and explain what we want. In our case, we are going to create one called "Explain it simply." The idea is that you can give it a report, some notes, or technical text, and it explains it to you in clear words and examples, keeping the important information. To create it, we are going to raise the reasoning effort to very high. We want it to analyze the examples well and extract some useful instructions. And once created, we can always set it back to medium for general use. The next step is to attach some examples of how we want it to transform the texts. Here I have

8:40prepared three documents. Each one has an original text and a simpler explanation below it. Each document explains a different concept and shows how to go from a more technical or dense text to a simple explanation, with clear words and easy-to-understand examples, which is what we are looking for. This way, we give it concrete

8:58references of how we want it to work. We want it to learn how to present the main idea, clarify difficult terms, use examples when it helps, and above all, simplify without changing what the document says. For this case, we don't need to give it access to an entire folder. We are going to use only these three examples we have selected. Now we complete the prompt. I will also leave it in the description in case you want to try it out. We now have the skill, and to use it, we can go to a task, type a forward slash, and search for it by name. We select it and tell it what

9:28we want it to do. For example, let's go back to the previous report where we have the business analysis. Now we can ask it to explain it to us for someone who doesn't know about business management. So, we select the skill and I tell it, "Explain the report to me for someone who doesn't know about business management.""Focus on the three money problems and what should be done first." This way, it uses the report already in this conversation and

9:50applies the instructions we just saved. Notice how it explains it now. A client who hasn't paid, a job we've delivered without invoicing, and an invoice we've paid twice. Keep the amounts, differentiate the three problems, and tell us what to do first. It's understandable without having to know about business management. And we can reuse that skill with other documents. We create it once, adjust what is needed, and we have a workflow we can repeat without writing all the instructions again. Another possibility with this combination is creating applications without needing to know how to program. If you use a simple tool that you pay for every month, you

10:31can try asking it to build a version with the features you actually need. In some cases, a custom tool for that job might work for you. And we can also make games. I'm going to use one for this demonstration because the result looks much better on screen, but the idea and the process also work for other types of applications. The first thing we do is change the project folder. I've created another folder called Game App by linking an empty folder to it. This is important because to create the game, it needs to generate files, and those files will be inside the folder we just chose. Now

11:05let's press the dictate button and explain how we want the game to be. Can you create a game for Windows that opens in its own desktop window and is called Orbital Rescue? I want to steer a ship with the arrows, rescue three small robots, and take them to a base while dodging drones and obstacles, with energy, a restart button, and a space aesthetic. Create it, open it, and test that it's playable. We send it and let it build it. Here we can follow how it's working. First, it prepares the game, and then it can open it and use it to verify that it works. When it

11:41needs permission to use a computer application, the corresponding request will appear. We review what it wants to open and allow it so it can continue. This is one of the interesting parts of the process. In addition to writing the code, it can start what it's building and interact with it to look for bugs. Astra is designed for this type of task that combines reasoning, tools, and computer use. Take a look at what it's created. Visually, it's very impressive and we have it open in its own Windows window. Let's test it. We steer the ship with the arrows and have to get close to these robots to activate the

12:13rescue. Now we head back to the base, dodging the obstacles. There it is, one out of three saved. Now we have the robot counter and the ship's energy. Let's go for another one to see how they respond while we dodge the drones. And as we can see, it works perfectly. We already have a first version we can play with. And now we could keep adding things, like, for example, a speed boost when pressing space with a more intense trail and a few seconds of cooldown between uses. But to ask for that change, I'm going to show you another feature of Work. Let's open a new conversation, and this time we'll

12:52use the live voice. This is getting quite close to the idea of having your own Jarvis. You talk to it and you can coordinate the work you have open. Earlier, we used the microphone to turn a request into text. Well, here we are going to have a conversation and give it instructions on the tasks we've already created. Hello, can you hear me

13:11? Yes, I hear you perfectly. In the other task where we created orbital rescue, I want you to add a speed boost when pressing space with a more intense trail and a few seconds of cooldown between uses. Pass that change to the game task. And in the conversation where we analyzed Lumbre Studio, use the report to draft an email for Irene with the main problems and what we should do first. Leave it as a draft in that conversation without

13:36sending it. Sure, let me take care of that. The voice here acts as a coordinator. I explain what I want and it passes it on to the other tasks. Look, the request for the speed boost has already arrived , and now this task can work on the change. And here we have the email task for Irene using the report we had prepared. And we also see that the game has updated, and now pressing space gives it that speed boost. This way, you can speak from a single conversation to give instructions to several tasks and check how they are going. By the way, in the voice conversation, use GPT Live. From it, we

14:11are coordinating the tasks we have opened with Astra. The interesting thing is that you don't have to go into each conversation to write the changes separately. You ask for them by speaking and then open the results to review them. With several projects underway, this can be very convenient.

14:30And before we finish, let's talk for a moment about usage, because Astra can help you a lot with difficult jobs, but it can also use up your available limit quite quickly. That's why it's best to find a balance between what it provides and what it consumes. I wouldn't use it automatically for every task. The examples we've seen fit jobs where it's worth trying a more capable model, analyzing many files, studying a writing style, or building and reviewing an application. For other things, you can use a model from the 5.6 family. Think, for example, of a scheduled task that checks messages, prepares a news summary, or classifies

15:06information following very specific instructions. You might notice some difference when switching to Astra. The question is whether that difference is useful enough to offset the consumption . There is a simple way to check it. Take a task you do regularly and run it twice, once with Astra and once with the 5.6 model you are currently using. Give it the same files and the same

15:28instructions and compare the results. If you don't find a significant improvement, you can continue using the more economical model for that task and save Astra for work where it actually adds value. You can also check the usage section of your account. Consumption depends on the model, the reasoning, and the work you perform. Not all messages cost the same. So that is the GPT6 Astra combination within ChatGPT Work, an environment where you can now assign complete tasks to a much

15:56more capable model. M.