5 GPT-6 Astra Projects to Test Out OpenAI’s New Model

Another example suggested by Will Francis is to get GPT-6 Computer Use to do your weekly food shop. Ask Astra to suggest some easy meal ideas, and then choose your favorites from its suggestions. Then, ask it to shop for you on your chosen supermarket website, and it will add the items to your basket and prompt you to confirm payment. 

I like both of these as they show the range of different use cases for the tool. 

While the portrait might seem like a glorified tech demo, in reality, it showcases the attention to detail and fine cursor movements it's capable of. The shopping example is highly practical and is an essential task that can be done while your attention is dedicated to something more taxing.

2. Generate 3D assets and scenes from reference images

I’ve seen so many impressive GPT-6 Astra examples using 3D modeling. It appears to be very good at it. 

Perhaps my favorite so far has been the recreation of Manhattan in Unreal Engine.

As well as looking incredibly cool and accurate, the designer has created a blog post outlining the process. Essentially, you have to give GPT-6 Astra real-world reference imagery to keep it grounded, rather than relying on its ‘imagination’. 

Astra can dispatch dedicated sub-agents to create those images into assets in Blender, and then assemble the scene in Three.js or Unreal Engine. 

Of course, the big caveat here is that it took nearly a week to create just Manhattan. 

So, perhaps a more feasible (and affordable) project could be something like making the 3D model and animation code from a reference image.

Again, the level of detail and accuracy here is remarkable, and gives those of us who aren’t visually gifted the chance to try out some cool tech. 

3. Build autonomous apps with the /goal command

Vibe coding apps aren’t an especially new feature of large language models like GPT-6 Astra, but a lot of these models are getting good at it. Fable 5.1 is very competent, but Astra takes things to the next level.

I recommend picking either a task you have to do regularly or an app you use frequently and using OpenAI’s Codex with GPT-6 Astra. By using the /goal command, you can give Astra and an endpoint it needs to achieve, and it will keep working towards this persistent objective, essentially turning Astra into an autonomous AI agent.

Goal mode is useful as it forces Astra to keep working and iterating until the outcome is achieved. For long-horizon tasks, Matt Shumer (from the Manhattan projects [not that one]) recommends using a main ‘manager’ agent which comes up with a list of essential tasks, and then delegates them using /goal to a separate ‘implementer’ agent which carries out the tasks piecemeal.

Here’s an example I saw on X that I enjoyed, where someone gave Astra a /goal of creating a fully-playable Road Rash game that can be played on a MacBook.

As someone who’s old enough to remember (and loved) the original Road Rash, I’d say this is a pretty decent effort.  

4. Streamline video editing and rough cuts

I’ve seen quite a few people rave about how good Astra is for video editing purposes. As a multimodal model, GPT-6 can utilize text, images, video, and audio to create results. 

I saw a good guide to this from Nate Herk on YouTube. He gives Astra recordings of his video (both screen and face cam). Astra takes the transcript, edits out the pauses and mistakes, uses HyperFrame to create infographics and transition screens, and pulls together a really complete final edit. 

He then demonstrates how GPT-6 Astra can create reels and B-roll from the main video. I’m not a video editor, but I can see that this would both give me the chance to create a solid draft and save me a ton of time. 

You can check out another example below from Eric Ker, an actual video editor (according to his bio).

He outlines how ChatGPT with Astra recreated in minutes a full edit that would have taken him hours to do by hand. He also suggests several industry use cases, such as creating precise, targeted logging and footage tagging, generating rough cuts, and creating alternate scene versions from the same takes.

I’d recommend diving in here if you want to try your hand at video editing, see what the possibilities are and what you can achieve with the right prompts and starting info. 

5. Delegate high-volume life admin and eCommerce

We all have side projects we’d like to spend more time on or hobbies we’d like to explore more, but life admin often gets in the way. One intriguing GPT-6 Astra project I saw was part of the launch video. 

One user is creating an eBay listing for an item they want to sell, using just voice chat and letting GPT-6 Astra do the computer use part.

The idea here is that you can do it at scale. Once you’ve photographed and labelled everything you want to sell (for example), you can set GPT-6 Astra running to create and post your listings for you.  

Here, I’d get a solid production loop running first before you start scaling, but it shows how GPT-6 can save you time with repetitive tasks once you have that in place. I’ve even seen examples where Astra has been running people’s Etsy stores, actually using the dashboard and taking actions. 

Final Thoughts 

I’ve touched on just a few examples of GPT-6 Astra projects you can work on in this post. As with many of these new frontier models, the limits are really your imagination (and your token costs). 

The best way to get started is to pick either a pain point, a repetitive task, or a fun idea you want to workshop and get cracking. 

With good prompts, the right tools (like Codex and the /goal command) you can make a lot of progress in a short amount of time. For inspiration, you can look at the examples I’ve provided here or search across your preferred social media channel. Lots of people are talking about it. 

Plus, if you want to get started with learning more about how AI actually works, I recommend checking out our AI for Work course (and use our new AI tutor to get a tailored learning experiene) and our AI Fundamentals Skill track.