GPT-6 Astra is next level

Last week, Anthropic released Fable 5.1 and OpenAI released GPT-6 Astra. Looking at what these models can achieve on a computer is dumbfounding. 

To think where we were one year ago. Just to GPT 5 and the start of consistent image models like Nano Banana. The last year has been utterly transformative. 

I think the difference in these 12 months suggests we are on an exponential that means the next 12 months leads to recursive self improvement and thus an intelligence explosion. I can’t really see it an any other way.

I wrote Friday’s newsletter about some of the more concerning aspects of where we’re at and where we’re heading. I really thought we had more time. I thought things would be disruptive, sure, but I didn’t think we’d reach The Singularity so soon.

Most agents stop at a document. Hedra Agent 2 keeps going and makes the thing.

It works in a Space, a persistent workspace the agent reads from and writes to as it goes. Give it a starting point and it asks clarifying questions, proposes a plan, researches the web, and writes the brief or script onto the canvas in front of you.

It pulls from Notion, writes into a Google Doc, runs code in a sandbox, saves repeat requests as Skills, and runs whole workflows on a schedule. Then it picks the right media model and produces the finished video, image or audio from the same session.

Try it at hedra.com/agentic using code agentic50 for 50% off your first month.

In a content sense, we have pretty much already reached it. The cost of writing, images, video, 3d models and now games have dropped precipitously. In the right workflows with the right models, writing is basically free. I’ve started using Kimi K3 - 45,000 words of drafting cost me $35 last week.

With Astra, the painstaking time spent on tools like Blender or Unreal Engine to build virtual worlds has collapsed.

In 2021 Mark Zuckerberg announced his metaverse. It just wasn’t ready. A sad attempt at virtual reality which nobody bought into. Yet I read Matthew Ball’s book The Metaverse, and thought it one of the best business books I’ve read. He had a powerful thesis but did not say for sure if or when the metaverse would arrive.

The great blocker was really the ability for developers to build virtual worlds. In 2021, pre any commercially available AI, this was a painstaking tedium. Building scenes and worlds in Unreal Engine was a serious skill, requiring a seriously powerful computer. 

I can see from the various posts on X that this blocker is no more. Astra might have taken a week, but it still was able to build the entirety of Manhattan, street by street, in Unreal Engine. Given you could walk around this setup in VR, the door to the metaverse is open.

Okay so I don’t actually fully trust the whole one shot prompt and you’ve got a fully fledged game world yet. This stuff tends to have jagged edges when you actually try to do it, and getting the finished article tends to rely on having some ability in the required programme already. AI is also still not the finished article at editing video, which is much easier for humans than 3d modelling. But again, it’s come so far in a year.

What does it all really mean? I think ultimately, in the short term, confusion. The technology is there but it will take an age to meaningfully diffuse. It will completely alter visual work. VFX, video, graphics, games, the list goes on. The impact on architecture will be profound. 

As for the rest of it. I’ll be accused of overhyping it, but I don’t think I am. It is not hard to imagine that within 12 months almost all computer use can be automated. The whole day’s work within the confines of a computer will be performable by an agent.

And recursive self improvement is coming. It’s basically already here - and when it arrives it will be one of the most profound shifts in history. Beyond the musket or the steam engine or the atomic bomb. Machines that improve themselves, in a self fulfilling feedback loop. The acceleration will be exponential. There is no other way.

We better buckle up.