Sep 29, 2026 · 2h 6m
Listen from 9:53
Listen at 9:53
OpenAI bets proactive agents will remake everyday computing
OpenAI DevDay | Sam Altman, Alexander Embiricos, Andrew Ambrosino, Peter Steinberger, Tibo Sottiaux, Kath Korevec, Casey Neistat
The episode captures a pivotal product shift as AI moves from answering prompts toward acting across software, devices, organizations, and creative work.
- 1OpenAI is moving beyond chatbots toward agents that can operate computers, remember context, coordinate work, and act proactively.
- 2Codex, DOTS, Spaces, and Sites show how agentic tools could spread from developers into businesses, teams, and ordinary users.
- 3The promise of autonomous assistance depends on approval boundaries, safety review, open ecosystems, and interfaces that make capability understandable.
Don't miss
Peter Steinberger describes how OpenClaw let users see agents install software, buy things, and act directly on their machines.
The brief
Recorded live at OpenAI Dev Day, TBPN traces AI’s move from conversational answers to agents that operate computers, coordinate work, retain context, and take initiative.
Sam Altman frames adoption as lagging capability, while arguing that mundane personal conveniences may matter more to users than spectacular scientific breakthroughs.
Alexander Embiricos and Andrew Ambrosino show how Codex, DOTS, ChatGPT, and Space are converging into agentic tools for developers, teams, businesses, and solopreneurs.
Peter Steinberger pushes the idea further: agents could install software, buy things, modify hardware, and collaborate in swarms, raising questions about identity, ownership, and control.
Kath Korevec and Casey Neistat widen the frame from generated websites to physical game hardware, suggesting AI may reshape both software creation and the devices people use.
What was said on this episode
82 statements · 67 positive · 7 negative · 2 mixed · 6 neutral
Applying AI effectively can enable companies to increase headcount.
“if you're actually applying AI well, you can actually grow. Sure. Grow your payroll.”
Listen at 6:13
World models could eventually be licensed as game-development infrastructure.
“people pay Unreal Engine to, you know, host or develop their games on top of”
Listen at 8:23
AMD’s World Labs acquisition represents a larger future bet on robotics.
“this is like a future-facing bet as in, hey, let's make a much bigger play into robotics.”
Listen at 8:42
DOTS can restore users’ attention, time, and ability to explore.
“it is amazing for that. And I feel like it has gotten me back, like, my attention and time and ability to like explore and do new things.”
Listen at 10:17
AI can give people more time for family, hobbies, and creative projects.
“I think it's great to give people like more time to hang out with their kids, do their hobbies, work on creative new projects.”
Listen at 11:00
People will gain more time for family, hobbies, and creative projects.
“I think people are going to get that.”
Listen at 11:11
AI progress and societal change will accelerate over the next six months.
“probably the next 6 months will be even crazier.”
Listen at 11:42
Always-on agents will expand how people use AI in daily life.
“with these sort of always-on agents, I think people will again go through a period where they're like, wow, I can really use this in all kinds of ways.”
Listen at 11:58
Highly capable AI models will improve people’s lives in unforeseen ways.
“models that are this capable, are going to be applied to make people's lives better in big and small ways, in ways that we can't even really imagine today.”
Listen at 13:33
OpenAI intends to let users use ChatGPT subscriptions across other services.
“we want people to take, well, to bring whatever they want into ChatGPT, run other apps there, but also to take their subscription and go use it elsewhere.”
Listen at 14:41
OpenAI’s proactive-agent messaging is currently too conservative.
“I think we've tuned the dial a little too conservatively for now.”
Listen at 15:53
OpenAI’s chip program will begin operating in early 2027 and expand later.
“our chip program, which will start to come online in the first half of 2027 and will scale a lot in future years.”
Listen at 18:34
OpenAI’s chip program will provide a major inference advantage.
“I think gives us a huge inference advantage”
Listen at 18:41
A single AI model could automatically choose appropriate capabilities and trade-offs.
“the dream is that there is a single model and it kind of makes great decisions for you.”
Listen at 19:35
Scientific discovery will accelerate substantially because of AI.
“I think we'll be very right about the rate of scientific discovery.”
Listen at 21:03
Many aspects of human life will remain broadly unchanged despite advanced AI.
“a lot of parts of life are going to stay the same.”
Listen at 21:18
AI lets developers build and test many ideas before concentrating effort.
“You can now build all 30 things and you can get people to use them all and then you can work more on the ones that work.”
Listen at 22:11
AI alignment science has not been solved.
“Anyone who says we have solved the science of alignment, I believe, is wrong in a very dangerous way.”
Listen at 22:45
Developing superhuman AI requires solving the science of alignment.
“we have to actually solve the science of alignment.”
Listen at 23:10
DOTS can handle delegated tasks for users.
“If you want to delegate it, sure, go use a dot.”
Listen at 26:34
A DOT can operate a cloud computer and perform computer-based tasks.
“your dot has its own computer, can do anything that you can do on a computer in the cloud.”
Listen at 27:19
DOTS can autonomously build video games.
“we've just seen people having Dots just like autonomously building like really fun games and stuff.”
Listen at 29:39
DOTS can manage operational tasks for solopreneurs and small businesses.
“having a Dot that just like has your back and is like managing all this stuff for you”
Listen at 30:02
Specialized DOTS operate using their own computers.
“specialized DOTS have their own computer, just like DOTS.”
Listen at 32:06
Organizations can deploy specialist bots to accelerate functional work.
“if you are a team or an organization and you have a specific type of like functional work that you would like to like accelerate the whole team with, yeah, that's where you deploy a specialist bot.”
Listen at 32:22
OpenAI’s agents will use identities distinct from their users.
“we're going to have like a strongly separate identity. They're going to look different.”
Listen at 34:03
AI makes HTML authoring nearly costless.
“HTML authoring is like basically free.”
Listen at 35:36
AI can create interactive artifacts and games inside documents or pages.
“you could just be like, hey, like build me this artifact that I can click around or like literally put a game in the doc or whatever in the page.”
Listen at 35:43
DOTS include safety protections and customizable rules.
“DOTS have many layers of like safety protections and one of them is custom rules”
Listen at 36:34
ChatGPT users can install plugins that render custom interfaces inside the app.
“if you're a ChatGPT user, you can install a plugin that is a plugin extension that actually renders like custom UI, custom pages just in the ChatGPT app.”
Listen at 37:32
AI interfaces should adapt to users’ preferred contexts and modalities.
“the interface should just be whatever you want.”
Listen at 38:38
Voice interaction with DOTS can be highly effective and enjoyable.
“I really love talking to my Dot over voice.”
Listen at 39:17
Proactive agents can detect needs and research solutions without explicit prompts.
“it'll just notice that someone was urgently asking me for the headshot. And so it might just go ahead and proactively research like, where is the headshot?”
Listen at 41:18
Proactive agents will sharply increase the value users obtain from existing models.
“we're about to achieve this like step function change in how much value people are getting from existing models.”
Listen at 41:32
ChatGPT Work now operates by default across desktop and cloud environments.
“now by default ChatGPT work on the desktop.”
Listen at 43:46
OpenAI’s new products run in the cloud while accessing users’ local machines when needed.
“all of our new products today, whether that's the Dots, the Codex Cloud, ChatGPT Work, all of this stuff is running in the cloud and has access to your local machine if it needs it.”
Listen at 44:19
Most newly released OpenAI products are iterations of earlier experiments.
“most of the stuff that we released today is like a take 5.”
Listen at 46:55
OpenAI’s newly released products currently feel highly impressive.
“the stuff that we have out today feels, feels magical.”
Listen at 47:54
Space moves Codex and ChatGPT into a more collaborative, multiplayer workflow.
“this is really the entry of Codex and ChatGPT into a multiplayer, more collaborative era.”
Listen at 48:42
UltraFast substantially improves workflow speed once adopted.
“UltraFast has been one of those things that for us, like, you just, you can't go back”
Listen at 51:46
AI adoption is spreading from engineering and research into other organizational functions.
“engineering was here, research is here, and every other one is like quickly following in sync.”
Listen at 53:24
AI agents can install software, make purchases, and perform computer tasks with suitable tools.
“it can install stuff on your computer or buy things for you or do whatever you want that you can do on your computer if you just only give it the right tools.”
Listen at 57:57
AI agents should proactively act rather than wait for user prompts.
“what about if your AI could actually just be proactive?”
Listen at 58:32
OpenClaw’s capabilities were limited by model and software quality.
“there's still limitations Models were not as good. The software was not as good.”
Listen at 1:00:20
Steinberger favors one primary agent but sees a future of multiple agents as plausible.
“I'm more in the first team, but I can also totally see like the the second team”
Listen at 1:02:29
Users should be able to bring their own agents to external services.
“I want my agent to be able to use your system in the way it thinks it's best.”
Listen at 1:05:42
Fastest-click competition may no longer suit agent-mediated transactions.
“Maybe the who clicks fast enough is no longer the right way.”
Listen at 1:06:54
Transaction systems should be redesigned for agent-mediated interactions.
“there's a lot of ways where we just need to rethink how we did things in the past because it no longer makes sense.”
Listen at 1:07:10
Personal agents should continuously think and prepare useful assistance.
“I want an agent that's always thinking.”
Listen at 1:07:49
Future interactions will involve agents on both sides of transactions.
“any worldview that assumes that only one side of the interaction will be agent-mediated is mistaken.”
Listen at 1:09:13
AI systems will increasingly compete against other AI systems.
“You're going to fight AI with AI.”
Listen at 1:09:38
Improved AI agents make software and hardware stacks increasingly modifiable by individuals.
“I can now literally own the stack and there's nothing we cannot fix anymore.”
Listen at 1:13:34
Future AI advances could make building a personal operating system feasible.
“with enough tokens and like advances, that ridiculous thing will not be that ridiculous anymore in the future.”
Listen at 1:13:48
Products for broad audiences require different construction approaches than smaller-scale systems.
“Not everything works in that scale to that scale. It's a very different way how you need to build things.”
Listen at 1:15:41
Developers are especially excited about ecosystem openings such as Sign in with ChatGPT.
“mostly ecosystem and, you know, us opening up like sign in with ChatGPT”
Listen at 1:16:53
OpenAI ecosystem access will make developer product experiences more cost efficient.
“you can do that in a way that will be much, much more cost efficient.”
Listen at 1:17:33
DOTS remains continuously available and proactively researches users’ goals and preferences.
“the agent never stops. Occasionally it sleeps, but it's always online, always doing proactive research, thinking about your goals, thinking about your preferences.”
Listen at 1:18:44
Teaching DOTS user preferences improves its performance.
“the more you interact with it and the more you teach it about your preferences, the better a job it does.”
Listen at 1:19:01
OpenAI plans to bring DOTS-like technology and capabilities to 1.2 billion ChatGPT users.
“in the future, we'll ship a lot of this, the same technology and the capabilities that you read in ChatGPT as well for our 1.2 billion users.”
Listen at 1:20:23
Agent harnesses are generally built ahead of model capabilities to compensate for limitations.
“you always build the harness a little bit ahead of the capabilities of the model in order to kind of make up for it.”
Listen at 1:21:11
The Codex harness powers DOTS.
“The Codex harness is used to power DOTS.”
Listen at 1:21:34
DOTS agents have capabilities exceeding PhD-level performance.
“DOTS are like beyond PhD level, right?”
Listen at 1:22:53
DOTS bots can operate computers similarly to human users.
“Your bot can just like use a computer just like you would.”
Listen at 1:23:01
Computer-use performance improved tenfold over one year.
“I do think we have seen a 10x over one year speedup.”
Listen at 1:24:15
Guardian monitors primary agents to review sensitive computer-use actions.
“we have a second agent, which we call auto review, or internally we call it Guardian, which watches over this primary agent.”
Listen at 1:24:48
The Decisions API can produce approximately one decision every 200 milliseconds.
“you can make one decision roughly every 200 milliseconds.”
Listen at 1:26:50
Real-time agent actions will emerge within a year and beat OpenAI’s original Dota model.
“we'll see like, you know, definitely like real-time, um, real-time actions, uh, within the next year and like beating that original model.”
Listen at 1:27:01
AI systems should automatically allocate resources according to users’ goals.
“the system should adapt and, you know, configure exactly like how many resources it should use in order to achieve your goal.”
Listen at 1:28:28
DOTS will eventually make model pickers seem unnecessary.
“eventually it will be so good that you will kind of be shocked that we had a model picker ever.”
Listen at 1:28:55
Space provides shared organizational memory across an organization.
“Space is, you know, like a shared memory system that you maintain across the organization.”
Listen at 1:32:05
ChatGPT Sites launched plugins enabling integrations.
“we just launched plugins in Sites today.”
Listen at 1:38:19
ChatGPT Sites are integrated into Space.
“Sites are in Space.”
Listen at 1:39:05
OpenAI intends to make ChatGPT Sites available to everyone.
“we are very, very interested in making it available to everybody.”
Listen at 1:40:31
The Codex plugin can flash vibe-coded games directly to Chromatic cartridges.
“you can like, you can flash it to this cartridge immediately.”
Listen at 1:43:54
Retro gaming constraints encourage parents to physically play with children.
“it forces you, because of those constraints, like, physically sit with my kids.”
Listen at 1:45:58
Technology is beginning to return toward single-purpose devices.
“I think that we're starting to see like technology start to go full circle.”
Listen at 1:47:17
Consumer hardware will change dramatically over the next decade.
“all the hardware that we use is going to change dramatically in a way over the next 10 years that it didn't in the last 10 years.”
Listen at 1:48:45
Modern LG TVs continuously listen through their microphones.
“the microphone that was intended for like, change it to HBO, yeah, it's actually listening all the time.”
Listen at 1:53:39
LG TVs store microphone audio offline and upload it after reconnecting.
“when it wasn't networked, it wasn't connected to the network, it would store the audio files locally, and then when it reconnected, it would upload it.”
Listen at 1:53:46
A fully AI-generated work will eventually be undeniably good.
“I think it's gonna get to a point where there's gonna be something that's made on AI that's 100% AI, that's undeniably good.”
Listen at 1:57:45
Lower creation costs increase both low-quality and high-quality output.
“when you open the top of the funnel that wide, the amount of shit being made increases. But so does the amount of quality stuff being made.”
Listen at 1:58:12
Human oversight can make fully AI-generated documentaries factually correct.
“these 100% AI-generated that clearly have a human editorial oversight because they're factually correct.”
Listen at 2:01:44
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.