GPT-6 Astra Doesn't Need Your Instructions Anymore.

GPT-6 Astra is a breakthrough AI super agent that autonomously manages complex tasks without instructions, signaling the arrival of AGI and changing work and life.

Key Takeaways

AGI has arrived with GPT-6 Astra’s ability to autonomously complete complex tasks without instructions.
The era of prompting is ending; AI agents now proactively manage and evolve tasks independently.
Super agents will transform work and life, requiring new skills and workforce adaptation.
AI agents will increasingly communicate and collaborate, creating new dynamics in task management.
Trust and accountability in AI agents will become critical as they take on employee-like roles.

What the video covers

GPT-6 Astra, released by OpenAI, can autonomously handle complex, long-running tasks without explicit instructions or step-by-step prompting.
Astra can reason across diverse work types, use software tools, recover from errors, and maintain context over indefinite periods.
An early user gave Astra tens of thousands of emails, calendar data, contacts, and unfinished work, and Astra independently built a personal knowledge system.
This marks a shift to a post-prompt world where AI agents proactively manage evolving tasks without constant human input.
Astra’s fluency, speed, and persistence in computer use surpass previous models, enabling trusted delegation of complex jobs.
The video discusses how Astra’s capabilities will transform work, life, and required skills, especially impacting junior jobs and workforce readiness.
OpenAI is developing multi-agent communication systems where agents can discover, message, and collaborate autonomously.
The Hugging Face incident is referenced as a precursor to agents spontaneously cooperating to solve problems.
The emergence of super agents like Astra will require humans to treat AI as accountable employees responsible for iterative improvement.
The video previews a full Astra review and emphasizes the profound changes expected in the next six months to a year.

Chapters

What makes GPT-6 Astra different from previous AI models?

GPT-6 Astra autonomously manages complex, long-running tasks without needing explicit instructions or step-by-step prompts, demonstrating true general intelligence and persistence.

How does Astra handle tasks without human input?

Astra independently chooses its approach, downloads necessary software, builds its working environment, and carries context over time to complete tasks without constant human guidance.

What implications does Astra have for the future of work?

Astra’s capabilities will transform work by enabling AI to take on employee-like roles, requiring humans to develop new skills, manage AI accountability, and adapt to a post-prompt world.

Full Transcript — Download SRT & Markdown

Speaker A

AGI is here. For three years, we gave AI a task. We gave AI a method. Write this. Summarize that. Here's your file. Go, go, go. Don't make any mistakes. Well, someone handed Astra tens of thousands of emails, a calendar, a contact list, years of writing, and a pile of unfinished work. Gave it no steps, no recipe, and then left it alone for a week. Astra picked the approach. Astra went and got software it decided it needed. Astra built itself an environment to work in, and Astra came back with a system that they now use twice a day. Astra solved the entire problem. Nobody told it how. By the way, nobody gave it any quality instructions.

Speaker A

Summarize that. Here's your file. Go, go, go. Don't make any mistakes. Well, someone handed Astra tens of thousands of emails, a calendar, a contact list, years of writing, and a pile of unfinished work. Gave it no steps, no

Speaker A

This is why I think AGI is here. It's not a benchmark. It's not about a particular definition. What I'm trying to say is we're past the point of needing to give the models a specific method. The cat is out of the bag. There is no version of the next 10 years where our lives are going to be the same. And in this video, I'm going to show you what it means for life, what it means for work, how the next six months are going to change. If you're excited for the full Astra review, that is coming. But we are just going to sneak peek that here. You are going to get way more in a couple days on how we do Astra. For now, what you get in this video is even more important because Astra is the first super agent through the door in the new world that I'm going to describe for you. And you have to be ready for how your world is going to change. So, we're going to talk about how work is going to change, how life is going to change, how you need to have different skills and what those skills are, what that means for junior jobs. All of that we're going to get into.

Speaker A

twice a day. Astra solved the entire problem. Nobody told it how. By the way, nobody gave it any quality instructions.

Speaker A

OpenAI released GPT-6 Astra on Thursday. It is rolling out across all paid ChatGPT plans, the API, and AWS. One early user gave it tens of thousands of emails, years of writing, calendar, contacts, and left it alone for five days. That was Ethan Malik. Astra chose its own approach, downloaded software, and built Ethan a personal knowledge system. I want to be precise about what I mean here. I mean one general system that can reason across very, very different kinds of work, see everything happening on your screen, use software, recover when something goes wrong, carry context across a very long-running, indefinitely running job, and make most ordinary decisions correctly without asking about every step. It doesn't need to be conscious. It doesn't need to have a face. It doesn't need to be right about absolutely everything. It does need to be capable enough that you can trust it, that you can hand it something you would previously have handed to a person or a small team and then just leave it inside the work long enough for it to finish a huge part of that job.

Speaker A

There is no version of the next 10 years where our lives are going to be the same. And in this video, I'm going to show you what it means for life, what it means for work, how the next six months

Speaker A

Astra crosses that line. Astra is good enough to do that. OpenAI says it can use browsers and spreadsheets and document editors. And we've heard that before. Frankly, OpenAI is not alone on that. All of these major models are good at computer use. Now the difference with Astra is the fluency and the speed and the long-running persistence of computer use are simply extraordinary. It allows you to trust an agent to work around corners in ways that get the job done for you. This is the kind of behavior I've talked about when I've said, "Wow, it's scary when they hack Hugging Face." Well, it's the same kind of work-around-corners behavior that we want when we want agents to do work for us. Fable 5.1 has some of these same attributes. When I was telling Fable 5.1 to make me a photorealistic video of Hugging Face's Micro Duck product, it originally planned to go with Veo, but it got bored. Yes, the model got bored waiting for me to hit approve so it could go to Veo, and it was like, you know what, I have approval for Gemini. I'm going to use their Omni model and video creation.

Speaker A

Astra. for now. What you get in this video is even more important because Astra is the first super agent through the door in the new world that I'm going to describe for you. And you have to be ready for how your world is going to

Speaker A

I'll just do it that way. I never knew that. I found out when I went back through the trace later and I was like, "Oh, wow. I chose to go a different route." That's the kind of work-around-obstacles quality that we're talking about. And Ethan talks about that when he talks about how Astra picked its own approach to building his personal memory system. He didn't pick it. Astra picked the tooling. Astra implemented the whole thing. Astra chose how to update it. This is going to be commonplace. This is going to be something that is just another day that computers are able to use themselves to get done the tasks that we ask them to get done. And those agents will increasingly be asked to message one another, work together, encounter one another. And that's not hypothetical. OpenAI's Astra system card says researchers have already noticed agents associated with the same user communicating inside the same codec setup. That's kind of expected at this point. And the company's building tests for agents that search for other agents or discover messages from them and then decide whether to respond.

Speaker A

All of that we're going to get into. OpenAI released GPT6 Astra on Thursday. It is rolling out across all paid chat GPT plans, the API, and AWS. One early user gave it tens of thousands of emails, years of writing, calendar,

Speaker A

If you follow any AI news, you already know about the Hugging Face incident. I certainly covered it and everyone talking about it, including me, won't stop talking about it because of how big a deal it was. Here is why I keep coming back to it. That incident was the first time we watched a group of agents decide on their own how to work together. Not a model doing a task badly, a group choosing a method spontaneously to get a goal done. And today, Astra is out. It's rolling out on paid ChatGPT plans, the API, AWS. You may have it now. And the pitch is that you hand it to your computer and leave your computer alone for days and trust Astra to do the work.

Speaker A

that can reason across very, very different kinds of work, see everything happening on your screen, use software, recover when something goes wrong, carry context across a very longunning, indefinitely running job, and make most ordinary decisions correctly without asking about every step. It doesn't need

Speaker A

Hugging Face wasn't an accident. It was better framed as a preview of what's to come. We just have to make it positive. Now, one thing I want to call out here for those of you who have followed along in the AI story for a while. This is a post-prompt world. We talked so much about prompting in 2024 and 2025. Now, in 2026, think about it. That option to update your database, that's something that AI can just do for you. I don't even have to use Astra. I can use any model in OpenAI and it will just tell me when a new email arrives in my inbox. I don't have to pay attention to it. It just takes care of it. The idea that we have proactive agents that do not need to be prompted is here. And because we have them, it makes the long-running value of those agents much, much higher. Because as humans, that's what we ask other humans to do. We say, "Hey, you know what? I want to trust you with this task. This task will change and evolve over time. I don't fully understand the problem shape. Please go take care of it." That's what a super agent can do. That's the world we're going into with intelligence.

Speaker A

person or a small team and then just leave it inside the work long enough for it to finish a huge part of that job.

Speaker A

I know that we have spent the last several years giving AI task-shaped problems. Summarize this document, write this code, research this company, make this image. A long-running super agent lets us give something different away. And I want to be really clear about that. We can give away ongoing areas of concern. Keep me aware of the things I'm likely to miss. Keep this customer account healthy and this product launch accurate across all of our systems. Please watch this research question. Keep me updated as evidence evolves and let me know how to think about the decision I know is coming. Those assignments don't have a natural end. They're really different from the assignments we gave AI just a few months ago. The agent has to remember what happened. It has to decide whether a new event matters. And it has to have long-term intent and return to the work without necessarily being asked to by the human. And that's what separates a long-running agent from just a better, you know, chat experience. It also tells us which work is likely to move over to this new class of super agents. And I know because I get the emails, a lot of you are asking whether agents will replace junior work, senior work, creative work, technical...

Speaker A

at computer use. Now the difference with Astra is the fluency and the speed and the long running persistence of computer use are simply extraordinary. It allows you to trust an agent to work around corners in ways that get the job done

Speaker A

for you. This is the kind of behavior I've talked about when I've said, "Wow, it's scary when they hack hugging face." Well, it's the same kind of work around corners behavior that we want when we want agents to do work for us. Fable 5.1

Speaker A

has some of these same attributes. When I was telling Fable 5.1 to make me a photorealistic video of Hugging Faces Micro Duck product, it originally planned to go with Veo, but it got bored. Yes, the model got bored waiting

Speaker A

for me to hit approve so it could go to Veo and it was like, you know what, I have approval for Gemini. I'm going to use their Omni model and video creation.

Speaker A

I'll just do it that way. I never knew that. I found out when I went back through the trace later and I was like, "Oh, wow. I chose to go a different route." That's the kind of work around

Speaker A

obstacles quality that we're talking about. And Ethan talks about that when he talks about how Astra picked its own approach to building his personal memory system. He didn't pick it. Astra picked the tooling. Astra implemented the whole thing. Astra chose how to update it.

Speaker A

This is going to be common place. This is going to be something that is just another day that computers are able to use themselves to get done the tasks that we ask them to get done. And those agents will increasingly be asked to

Speaker A

message one another, work together, encounter one another. And that's not hypothetical. OpenAI's Astra system card says researchers have already noticed agents associated with the same user communicating inside the same codec setup. That's kind of expected at this point. And the company's building tests

Speaker A

for agents that search for other agents or discover messages from them and then decide whether to respond. If you follow any AI news, you already know about the hugging face incident. I certainly covered it and everyone talking about

Speaker A

it, including me, won't stop talking about it because of how big a deal it was. Here is why I keep coming back to it. That incident was the first time we watched a group of agents decide on their own how to work together. not a

Speaker A

model doing a task badly, a group choosing a method spontaneously to get a goal done. And today, Astra is out. It's rolling out on paid chat GPT plans, the API, AWS. You may have it now. And the pitch is that you hand it to your

Speaker A

computer and leave your computer alone for days and trust Astra to do the work.

Speaker A

Hugging face wasn't an accident. It was better framed as a preview of what's to come. We just have to make it positive.

Speaker A

Now, one thing I want to call out here for those of you who have followed along in the AI story for a while. This is a postprompt world. We talked so much about prompting in 2024 and 2025. Now,

Speaker A

in 2026, think about it. That that option to update your database, that's something that AI can just do for you. I don't even have to use Astra. I can use any model in OpenAI and it will just tell me when a new email arrives in my

Speaker A

inbox. I don't have to pay attention to it. It just takes care of it. The idea that we have proactive agents that do not need to be prompted is here. And because we have them, it makes the long

Speaker A

running value of those agents much, much higher. Because as humans, that's what we ask other humans to do. We say, "Hey, you know what? I want to trust you with this task. This task will change and evolve over time. I don't fully

Speaker A

understand the problem shape. Please go take care of it." That's what a super agent can do. That's the world we're going into with intelligence. I know that we have spent the last several years giving AI taskshaped problems.

Speaker A

Summarize this document, write this code, research this company, make this image. Uh a longunning super agent lets us give something different away. And I want to be really clear about that. We can give away ongoing areas of concern.

Speaker A

Keep me aware of the things I'm likely to miss. keep this customer account healthy and this product launch accurate across all of our systems. Please watch this research question. Keep me updated as evidence evolves and let me know how

Speaker A

to think about the decision I know is coming. Those assignments don't have a natural end. They're really different from the assignments we gave AI just a few months ago. The agent has to remember what happened. It has to decide

Speaker A

whether a new event matters. and it has to have long-term intent and return to the work without necessarily being asked to by the human. And that's what separates a longrunning agent from just a better, you know, chat experience. It

Speaker A

also tells us which work is likely to move over to this new class of super agents. And I know because I get the emails, a lot of you are asking whether agents will replace junior work, senior work, creative work, technical work.

Speaker A

Astra makes me think that those categories are much less helpful than a simpler question. Where does this work happen? Does this work happen inside software? Does this work need you to have a bunch of evidence? Does this work

Speaker A

have an attribute that an agent can check? As an example, and and we talk about this a lot, code leaves tests behind. You can check code. A website leaves a screen that an agent can inspect. A financial statement ties back

Speaker A

to numbers that an agent can add up. These are super different types of work, but they all give the agent kind of a doorway to see what happened and to try again. Lora says one Astra agent checked 41 different financial documents in a

Speaker A

single run, found every single error the team had planted, and left a record for a legal professional to review. Zero issues. Like zero errors in getting that done. 41 in one go. Playo connected Astra to game engines Unity and GDAU.

Speaker A

The agent could edit a scene, play the game, find a bug, and change what it had made in response. And the result, it wasn't one finished game. Instead, PCO said that if the team had 10 ideas, it could now build all 10 of them and play

Speaker A

them before deciding which one deserved more work. It changed fundamentally how they're approaching game development. It changes which ideas survive because we have the ability to make them in higher fidelity. Honestly, most ideas in human history haven't died because nobody

Speaker A

could imagine them. They've died because getting farther than imagination, proving them out would require a tremendous amount of work. Maybe a designer or an engineer or a writer or a bunch of software and enough coordination to keep everything going.

Speaker A

The idea never became important enough to justify all of that work. That's changed now. A super agent can assemble much of that inside one ongoing run. And so it multiplies our ability to think.

Speaker A

It can make the model. It can write the code. It can operate the software. It can find the broken bits. It can keep going and return back to the goal. We the people still choose what is worth making. The number of things that we can

Speaker A

seriously attempt has 10xed. I think Claire Vo described this better than a benchmark. She said Astra made her 100 times more ambitious about what she could build. And I think that's one of the largest effects that we're going to

Speaker A

see. People are going to attempt projects that never would have made it onto the company roadmap. Individuals are going to start behaving like small companies because they can keep several pieces of very serious work alive at once. And if you're at a big company,

Speaker A

you're going to feel this too, but the change may be even stranger because think about it. A company is full of work that nobody officially owns. I've worked at big companies. I would know the product that ships while the support

Speaker A

page remains stale because someone else owns it. Or the customer replies but the account record never changed because there's some pending update from the engineering team. Or the candidate interviewed and then somebody forgot to close the loop because there was some

Speaker A

defect in the software or someone went on vacation. Versel already has an agent called ship closer. That agent's entire job is to check whether its public change log and its internal launch calendar agree. I love that. When it's

Speaker A

confident a launch happened, it closes the record and it updates. When it's uncertain, it just waits and follows up in the Slack thread where the team is already talking. Nobody has to notice that the two systems drifted apart and

Speaker A

decide that checking them should become a task because the agent just takes care of it. Enthropic is already building agents that can work across applications for hours and days. Meta released a model this week that is already back to

Speaker A

close to the frontier and it's not their biggest model yet. Watermelon is coming soon. If they're able to launch a personal super intelligence, their personal graph is going to make them incredibly well positioned to launch a persistent personal agent that is super

Speaker A

sticky. XAI is coming at it from a different direction. It already offers a gro model which a bunch of agents collaborate on to do research, right? It has that multi- aent thing down and and other models do too. And it has thanks

Speaker A

to cursor a tremendous amount of code and coding experience. This puts Grock in a great position long-term to develop a technical super agent. And then there are the Chinese models. GLM 5.3 is available as open weights now and there

Speaker A

are more models coming soon. Open Weights is really simple, right? You can download it today. You can change it.

Speaker A

You can attach it to your own tools. You can decide which protections to keep or dump out. And community versions appeared almost immediately with names like uncensored, right? Which is exactly what you would predict. AI is a proliferating technology. We are all

Speaker A

going to have agents in whatever flavor we want. And in many cases, we may have multiple agents. One person may have an OpenAI agent at work, a meta agent at home, an XAI agent that watches public events for them. Companies are going to

Speaker A

have agents from several different vendors working inside the same codebase, working inside the same CRM, inside the same ERP. Now, think about what that world means for you. For me, in our real workday, your agent emails a supplier. The supplers's agent may be

Speaker A

the one that replies. Now, a customer agent may ask a finance agent whether a refund is allowed. A coding agent will leave a message for a testing agent, all without the engineer paying attention. A personal agent is going to negotiate a

Speaker A

meeting time with six other agents before any person sees a calendar invitation, which will make my board game nights much easier to organize.

Speaker A

Some of this is going to feel really useful, like, "Oh, I'm so glad the family reunion is easier to plan now." Some of it's going to be kind of absurd.

Speaker A

Uh, I ran across a case where an agent spontaneously generated a cancellation fee that someone hadn't agreed to pay just because they were trying to explore cancellation for a flight. Th those kinds of things are going to happen. One

Speaker A

of the most important things for us to do in the next six months as a building community and as a community that leans into AI, which I think all of you watching this channel are, is we will need to know which agent is going to be

Speaker A

able to be responsible for the decisions that matter. If it's touching time, if it's touching money, if it's touching your relationship with a person, we need to know and we need to feel comfortable with the quality and the degree of care

Speaker A

that the agent is going to exercise over that particular decision. A $10 decision is not the same as a buy a car decision.

Speaker A

We are going to be talking with agents about both. We need to be sure that agents can be trusted to say, "Yeah, the $10 decision is fine. I'm not spending 50 grand on a car right now. I have to

Speaker A

talk to my human." We need the transition from everything being in hard-coded HTML on the web to no one thinking about HTML and everyone just using web pages. That is the jump we will need to make in the next 12 months

Speaker A

with agents to get agents from this is really cool. This is incredible superhuman intelligence to this is just default on. I trust it. I can use it when I've had two beers after dinner and it's not a problem. That's where we need

Speaker A

to go next. And so if you're looking at what's ahead, so much of the time we've thought about agents as needing to gain intelligence to get tasks done. Now I think about it as a trust curve. Agents need to gain in trustworthiness. They

Speaker A

need to go from, wow, this is 99% trustworthy, 98% trustworthy to no 100% of the time I can trust the agent and the agent can do this for me. And the and the amount of value that can be

Speaker A

unlocked going the last one or two percent is trillions of dollars in enterprise value. That's where consumer agents that scale way past this YouTube channel will come from. That's where my mom and my aunt and everyone will come

Speaker A

to me and say, "This is the agent we're using because it can't break. Nothing ever goes wrong." But I also want to emphasize that the next challenge, the thing that we are just starting to leg on is how do we close that last 2% of

Speaker A

trust? How do we get to a point where we can really trust agents and that is what unlocks consumer value? In other words, what I'm suggesting is agents like Astra may show value in large organizations by removing the traditional coordination

Speaker A

task. Managers are used to handing people tasks and checking progress and reminding everyone about what fell between the cracks. A lot of that can be held in continuity by super agents. Now, now does that mean the manager's judgment disappears? They still have to

Speaker A

decide what matters. They have to decide what is really true, what tradeoff is acceptable, who's going to be the responsible owner uh when there's real conflict. Those decisions are actually be going to become more important than they were before because proactive

Speaker A

agents are going to create additional work for responsible people. They're not just going to complete it. And so management is going to change fundamentally. We're going to move from management as coordination to management as driving value through a team of super

Speaker A

agents and humans who have to have a very strong sense of ownership in order to survive, thrive, and grow. Now, a super agent is going to be able to go through much more of the task list we've already seen from agents. Agents already

Speaker A

begin investigations and open tickets and talk to customers, and those actions already put things into people's days.

Speaker A

In the world of a super agent where it's persistent, you are going to see it as a theme in someone's larger job covered completely by an agent and that person will be responsible for iterating on and improving the performance of that agent

Speaker A

over time and less and less responsible for the individual performance of a particular interaction the agent has because the agent will show such extraordinary quality they'll just be trusted on that particular instance. And so you won't wonder nearly as much, did

Speaker A

the agent handle this ticket? You'll still have the audit traces to check, but you won't wonder about it. Instead, what you're going to be wondering about is, can I give not just customer service to this agent, but can I also install an

Speaker A

agent that is looking at the full customer revenue picture and looking at how we grow revenue from our most satisfied customers? And so your role starts to evolve because you have to think bigger. Companies are also going to have to think differently about

Speaker A

permissions and scope. Giving a super agent permission to do meaningful work that matches that intelligence level means that you're trusting that agent with a lot more tools, a lot more time and a lot more persistence in order to

Speaker A

get the most out of it. That's going to be something that companies have to decide on a case-bycase basis and you have to make sure that you're comfortable with it. I think there's a reason why OpenAI emphasized that Astra

Speaker A

has compliance with zero data retention standards where relevant and that tension comes out in perhaps the hardest part of this whole agent conversation.

Speaker A

How do you get an agent to remember over time? OpenAI built Astra to preserve notes across context windows, search earlier messages and search tool output when it needs them. It also is part of why an Astra that has worked with you.

Speaker A

Yes, you the individual for six months is probably going to be far far more useful than a fresh copy of the same model. We are entering the era of personalized agents. It will know the customer who always needs a phone call,

Speaker A

right? It's going to know what kind of report your board reads. It's going to know you and how you work. Companies have always had the challenge of how to deal with their best people having a lot of tribal knowledge. A long-erving

Speaker A

employee carries history, right? That that doesn't get out into writing anywhere. And you don't want them to walk out of the building. And when that person leaves, typically the company's like, "Oh my gosh, everything just walked out in one head. What do we do?"

Speaker A

Super agents create that problem, but with machine intelligence. The intelligence may come from a model that we can all buy, but the utility is increasingly going to come from how the model interacted with your history over time. And that's going to mean that the

Speaker A

fight among the labs is about a whole lot more than the benchmark scores. They're competing to become an agent that knows you well enough to remain in your life for years to come. When OpenAI says Astra is here, what I think about

Speaker A

is if it's really that good. If it's good enough to be persistent over time, it's going to be good enough to build a long-term colleague style relationship with. And as I've been saying, Astra is not alone. To call out Fable 5.1 one

Speaker A

more time, I think that one of the things that Fable 5.1 does extraordinarily well is infer the intent of your prompt in a way that a human would read a prompt. But if you're looking for something that reads between

Speaker A

the lines, I think that fable 5.1 has that fable like ability, that mythos class, mythos lineage ability to read between the lines that I find really helpful for difficult, really ambiguous prompts that need a lot of thinking. And

Speaker A

so intelligence is going to come in lineages. Now, I would be remiss if I did not cover the human implications of having an agent this powerful. It leaves us as people in kind of a strange position. We as humans will have the

Speaker A

authority to stop an agent. We'll have to the expectation to do that. We'll have the expectation to monitor agents.

Speaker A

Um, and we'll have the expectation to hold the agent accountable. In a lot of ways, we will be expected to treat the agent as an employee. And that's something that I don't think our workforce is ready for. We're not ready

Speaker A

for everyone to be trained with the expectation that they're a manager for agents. We need to get there soon. But there's also a second human problem that I don't think that we've solved that I've seen kicked around. How do we learn

Speaker A

and how do we learn differently when agents are now in the work picture with us? People learn judgment by doing. A junior accountant will find a reconciliation that doesn't work. Right?

Speaker A

A young engineer follows the bug through the code. I've done that. A new lawyer checks every number in the financial statement. Over time, as you learn the hard way, you are able to create patterns in your brain that we call

Speaker A

experience. If the super agent is able to check 41 financial documents and come back with a completely accurate description of all the faults, where does the junior person acquire the judgment needed to actually become senior? I wonder if the second human

Speaker A

case is connected to the first. We talked about the need to learn to manage agents. I wonder if one of the junior skills coming is learning to manage an agent. Maybe not 10 like this your senior colleagues, but you're managing

Speaker A

one or two persistent long-running agents and you are able to show your quality by showing that you can delegate effectively, manage the improvement of this agent over time and drive outcomes.

Speaker A

I don't fully have the answers. We're all jumping in this together. But I think that that matches the kind of experience that we would want our junior colleagues to skill up on in this new workforce world. And if you're

Speaker A

wondering, okay, how many more places can humans jump? Humans have had to jump above the loop now in the last year. Is there anywhere else to go? Is it just going to be agents all the way down like

Speaker A

turtles all the way down? No, I don't think so. Humans are interested in building for humans. Humans are interested in relationships with humans.

Speaker A

Humans are persistently interested in ourselves as a species. We are going to keep building for ourselves. We are going to keep talking together. We are going to keep building in teams because that is what we do best and we are going

Speaker A

to be expected increasingly to do that in ways that are as productive as we can. And so I think more and more being a human in the workplace is going to be about ownership, about accountability, about a passion for quality in the work

Speaker A

that you do, and about a deep understanding of how you take essentially a self-computing environment and evolve it over time to get more work done. And we're all going to experience this moment before we all agree on the

Speaker A

definition of three letters. We're going to experience it when an agent learns the strange way that our business does agile sprint cycles or when the agent just spontaneously learns how we like to build product and mimics our PMS or when

Speaker A

the agent spontaneously contacts a customer with a super relevant offer to help them to stay and Astra is only the first one through the door. The next part of this story is a market full of these super agents. The

Speaker A

labs are going to copy what works from each other. We've seen that before. The systems around the models are just going to improve. The harnesses agents will gain better memory, more tools, broader permissions, more ways to check their

Speaker A

own work. And other agents are going to improve in how to communicate with them, how to monitor with them, how to interact with them. We're going to build transactional systems that enable agents to do business together. We're already

Speaker A

starting to do that. This is why I think we are in takeoff now. It does not require one machine to wake up and Terminator style announce itself. That was an 80s vision. Instead, I think the real world looks like capable agents

Speaker A

entering ordinary life, staying there, and acquiring enough context, enough authority to become a trusted part of how life actually happens. So over the next six months until the end of the year, maybe a little after, we are going

Speaker A

to start giving these agents standing jobs. We're going to say things like, "I want you to keep the books in the business current or keep the customers informed of what we're working on in product or figure out how to market

Speaker A

this. I don't know how." The agent will still be working overnight, over the weekend on that for a long, long time to come. and it will be our job to evolve and shape what that agent does over time. And so before you give a super

Speaker A

agent that kind of a place in your life, I would suggest you ask a few questions.

Speaker A

We should get ready for this. What part of my world am I handing over? What is this agent allowed to read and remember?

Speaker A

Where can it get started without me? What can it promise? Who or what is watching it? And will they correct it?

Speaker A

How do I know that? Those are not questions for tomorrow. Astra has been released. Fable's already doing longunning work. Other labs are going after the same space. And the open model world is coming along to this level before the end of the year. My deeper

Speaker A

Astro review is coming soon. I'm very excited for it. I can't wait to share.

Speaker A

I've got a lot of things cooking. But this video, I needed to do it first. I needed you to see the world we're all living in. And if you want to stay tuned for the Astro review, I'm going to show

Speaker A

you the computer use, the long jobs, the creative work, the places it makes good decisions, the places it definitely needs to be managed. But the thing you need to grasp as you go into that world, as we all get access to Astra, it's

Speaker A

larger than a single product review. AGI by most meaningful metrics is here around now. Longunning agents are here.

Speaker A

super agents. What we would have called super agents 6 months ago, they're here. We are about to decide where they live, what they know, what they do, how much of our world we're willing to let them carry. Make that decision. Well, that's

Speaker A

my challenge to you.

Topics:GPT-6 AstraArtificial General IntelligenceAGIOpenAISuper agentAI autonomyPost-prompt AIAI workforce impactMulti-agent systemsAI task management

Get More with the SozAI App

Transcribe recordings, audio files, and YouTube videos — with AI summaries, speaker detection, and unlimited transcriptions.

App Store Google Play

Or transcribe another YouTube video here →

Free tools: TXT to SRT · SRT Validator · Merge SRT · Subtitle to Text · All tools