OpenAI Cooked with Dots, But What Will It Cost You?

Author: Theoretically Media · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: OpenAI Cooked with Dots, But What Will It Cost You? · Channel: Theoretically Media · 13:28 · Watch on YouTube

Topics: AI agents, Markets and economy, Privacy and data protection

Summary

OpenAI’s Dev Day centered on Dots, persistent agents that work in their own cloud computers and can be contacted through ChatGPT, Slack, or Teams. The presenter sees promise in Dots for substantial work, but stresses that they are automated bots rather than conscious agents, and notes that access is limited to paid plans with substantial pricing. Other announcements included a shared agent workspace, a new model and faster API mode, and privacy features. Comparing Dots with Meta’s Muse, the presenter finds Muse useful and efficient but raises concerns about its computer-access requirements and Meta’s reputation, framing the two products as competitors for the role of AI assistant.

Key points

Dots are persistent agents that can work in the background using their own cloud computer, browser, and scheduled jobs.
Users can communicate with Dots through ChatGPT, Slack, or Teams, including by voice; direct conversation does not count toward usage, but work launched in Codex or ChatGPT does.
The presenter describes Dots as bots with timers and jobs, not conscious AI, and says OpenAI presents them as an extension of the user.
OpenAI introduced Spaces, a collaborative area where users, teammates, and agents can work with shared information.
Dots were announced for paid plans including Pro, Business, Premium, and Enterprise, while the presenter discusses new $100, $200, and $500 plan tiers and changes to usage allowances.
The event also announced GPT 6.1, described as cheaper than Astra and comparable in intelligence, plus an ultra-fast mode offering 300 tokens per second.
The presenter tested Meta’s Muse with inbox and calendar access, reporting that it booked a DMV appointment and provided a useful news feed.
The presenter says Muse seems aimed more at casual users, while Dots are pitched for more demanding work; they also question Muse’s requirement for access to the full computer.
OpenAI previewed Private Intelligence, a business feature described as keeping user data from being retained or viewed by OpenAI staff.
The video frames OpenAI Dots and Meta Muse as part of a broader competition to become users’ AI executive assistant and companion.

Chapters

Start	Chapter	Summary
00:00	Dev Day and the arrival of Dots	The presenter introduces Dots as OpenAI’s headline announcement and explains how the persistent agents are intended to work.
03:04	Spaces, demos, and what Dots are	The video examines OpenAI’s collaborative Spaces feature, a Dots demonstration, and the agents’ cloud-computer setup.
06:09	Access, models, and pricing	The presenter covers Dots’ plan restrictions, GPT 6.1, ultra-fast mode, and OpenAI’s revised subscription tiers.
08:46	Trying Meta Muse	The presenter shares results from using Muse for calendar, inbox, and news tasks, along with concerns about privacy and access.
11:22	Privacy and the assistant race	The video discusses OpenAI’s privacy feature and compares Dots and Muse as competing AI assistants.
12:57	Closing thoughts	The presenter connects the emerging assistant ecosystem to a previous film project and signs off.

Quotes

did Open AI cook and if so, how much is that meal going to cost? — 00:00

these are bots with timers and jobs that can cross reference with one another. — 05:39

I had it book me an appointment at the DMV to renew my license. — 09:18

they’re previewing something called OpenAI private intelligence — 11:22

all of these companies fighting it out to become your new AI executive assistant and best friend. — 12:57

Concepts and entities

Dots, OpenAI, OpenClaw, Peter Steinberger, Muse, Meta, Space, OpenAI Private Intelligence, Jev, Opus 5.5

Full transcript (original language)

00:00 Well, OpenAI’s dev day just wrapped up and this was a pretty interesting one with persistent agents being the main theme here and clearly kind of their catch-up flex on the other always on always working systems out there like Meta’s Muse. There were some other announcements including ones that may or may not make everyone happy. But the big question here is did Open AI cook and if so, how much is that meal going to cost? Okay, dot your eyes and let’s go check it out. Kicking off the big news coming out of

00:30 dev day was the much anticipated and highly teased Dots. This was initially leaked as OpenAI’s AON and then later on as O before finally being revealed as DOTs. And considering that spooky season is coming up, I just want to remind everyone DOTs is they’re not good. It’s not good candy. It’s not. It’s just not. That said, Open AI’s dots are basically agents that work on your behalf 24/7 whether you’re at the computer or not and whether your codeex is open or not. Each dot gets its own like little mini computer or virtual machine uh somewhere

01:01 presumably on OpenAI servers uh and its own browser where it can write and test code. And of course you can message your dot or even hop on a voice call although we’ll look at that in a minute uh with it either you know obviously on mobile or in chat GP desktop or web I guess. Um, and of course you can message it through Slack or Teams. And that of course goes the other direction as well with your dot being able to message you or contact you via any of those methods as well. And yeah, I mean obviously if

01:32 all of this sounds familiar, it is a bit of a dressed up version of OpenClaw, considering that OpenClaw’s creator Peter Steinberger did join OpenAI back in February of this year, which is kind of funny if you think about it, considering that OpenClaw is now one of the partner apps. So you can spend your chat GPT’s paid usage on looking at the open source thing. We are going to talk about OpenAI pricing in a little bit as well. I mean obviously the pretty major release in this category recently has been Meta’s Muse which you know look despite anyone’s reservation about Meta

02:03 and privacy. You can’t deny this thing has been a quite a hit. It even ended up knocking chatbt off the number one spot in the app store at least in the US mostly because I don’t think it’s available everywhere else but that’s beside the point. We are going to talk about Muse a little bit later in the video because I’ve been running it for a few days now and I have thoughts. But in the meantime, where OpenAI seems to be differing from Muse is that they’re aiming for like more ambitious work. At least beyond book me a flight and order me those refrigerator filters that I know you haven’t changed in 3 years. And

02:34 not saying that with any kind of judgment. I haven’t either. Kind of a brutal disc. Sam called that kind of usage trivial. Uh and uh with their pitch obviously being that you hand dot like a whole boatload of responsibilities. the way you would give it to like a production manager with the idea being of course that your dot is chewing away behind the scenes continuing to work while you know your team is getting some muchneeded sleep. All of which is great. I do want to point out it is still not doing your laundry or mowing your lawn. We got to wait for the robots for that one. They also introduce space which is a pretty

03:04 interesting feature. Basically, this, you know, functions as OpenAI’s Google Drive. Although, uh, with some agentic space, creating, you know, pages where you and your teammates and your dot can all collaborate together. You know, I I I get it. I get what they’re trying to do here. It’s essentially putting up a walled garden around, you know, your information. To be honest, I’m not 100% sold on this. And I’m going to be curious to see how adoption works on all of this. I mean, obviously, it’s very clear what they’re doing. They’re not like throwing any shade at OpenAI here. They’re all doing it. uh you know

03:35 essentially trying to create walls around all of your data while simultaneously creating connections to everywhere that we actually are. So yeah, it’s going to be a really interesting carrot dangle to see uh how they can try to herd everybody into uh you know their their little uh magic data prisons. I mean we’ll see. Going back to the dotagentic and connectivity part of it, they did showcase a demo where a they were creating a fictional music platform launch through development, design, and release with their planned oh wow moment being a

04:05 voice call to dot. Although uh the voice response Holly, nice to hear from you. >> Hey, so I’m just walking in. Um can you catch me up on just user testing from last night? >> Checking that now. Don’t think that’s surprising to anyone who’s been using voice lately. Am I right? Overall, I mean, like, look, I mean, I’m being pragmatic about it, but

04:36 I do think that dots are pretty awesome. And it actually does legitimately bring us closer to like, you know, like you guys have seen the ridiculous Instagram and Tik Tok trends of like AI Jarvis, the guys who have like rooms that are absolutely drenched in LED lights and have like these insane like dashboards that do absolutely nothing and say things like, “Jarvis, make me $2 million in the next 24 hours. Make no mistakes.” I do want to shout out Eric Morrison’s brutal takedown of these clowns uh in his video AI LAR cringe lords. If you haven’t seen it, really awesome. Uh link down below. definitely go check it out

05:08 because I mean these guys are going to be going ham uh with dots. So, going back to dots, uh each dot will get what OpenAI calls its own cloud computer. We don’t know if like this is a Linux thing or if there’s it’s just a bunch of Chromebooks all like stacked together. Um but the important part being is that they are all going to be sandboxed from everyone else’s. And look, not to take anything away from like how remarkable all of this stuff is and how it works, but I mean, basically what this comes down to is that you have your own little mini computer somewhere on OpenAI

05:39 servers uh that essentially has hooks uh and cron jobs going on. And look, I I mean, I know that sounds reductive and I’m not trying to be at all. Um I’m just, you know, with the overall landscape the way it is. I just want to stress this is not like your conscious AI agent. Uh, these are bots with timers and jobs that can cross reference with one another. OpenAI’s own messaging is that they think of the dot more as an extension of you. So, uh, essentially when you talk to your dot, you are talking to yourself. Speak of which,

06:09 interestingly, um, just talking to your dot does not eat into your usage. That apparently is included in your plan. Although little tricky there because if the work kicks off in codeex or chat GPT work, it does count. And well, I mean, here comes the shoe drop is that DOTs are only available on the pro, business, premium, and enterprise plan. And they all supposedly uh, you know, get to start working with it in beta at as of today. Although, I will say that um I I am on the pro plan and my dot has not

06:41 appeared. That said, that’s fairly typical of any OpenAI rollout. I for whatever reason, I’m always the last one to get it. More on pricing in a minute and I you guys aren’t going to like it. Uh but the other big announcement was uh despite the teases of Astro 6.1 which we did not get mostly because it was deemed too powerful and they this news just broke the night before dev day. Um we did though get uh GPT 6.1 sold which obviously is you know way cheaper than Astra actually a fifth of the token

07:13 price uh and per open AAI near Astra intelligence and even smarter than Astra in some ways. That one should be available to everyone and I’m sure that we’re going to see benchmark charts any minute now. Uh they also introduced an ultra fast mode across the API chatbt and codeex. Uh this one is a speed tier 300 tokens a second which is which is kind of insane. Although ultraast I believe is only available via the API right now. So you know API pricing. Uh and and then now for the big news that

07:44 no one liked. Uh there is now a $500 plan uh pro 500 which is 25 times the amount of usage of uh the plus plan with ultraast included. So I’m sorry I I misspoke earlier. It is available via the API and the new $500 plan. Um they are bringing back the $200 uh pro tier and they’ve introduced a $100 pro tier. although the $200 plan is essentially usage-wise now the $100 plan. So

08:16 essentially you’re getting, you know, half the amount of usage out of it with OpenAI’s reasoning that the the models are more efficient now and that you’ll get more work done for less. We’ll see about that. In their defense, I will say that Opus 5.5 seems to be extremely token efficient uh in that, you know, it’s actually kind of hard to burn through your limit uh with Opus 5.5. So, I don’t know, maybe something maybe they’ve all kind of figured out something on that front. The other one that kind of made me chuckle is that uh there’s a there’s a new decisions API

08:46 that they previewed. Uh basically, it’s a super fast model that picks from like a short list of options essentially like yes, no, uh do this, do that, or like send this here or send this there. I mean, basically, it’s Jev. I mean, it is Jev, which I mean, I guess fair game. I never really got a chance to play with Jev. I never really found a reason to play with Jev. Now, conversely, in the same area, but on the meta side of things, uh yes, I have spent a few days playing around with Muse. Now, it is kind of an interesting contrast uh to OpenAI’s dots. Now, I decided to go a

09:18 bit yolo here despite like Claude’s like finger wagging. Uh and I did end up giving uh Muse access to my inbox and my calendar. And I mean, I do have to give Meta credit here. This is a pretty good product. Um I had it book me an appointment at the DMV to renew my license. And I mean, yeah, I mean, it took care of it. It It worked. Uh I mean, I still ended up in line at the DMV for 30 minutes, but that was not Muse’s fault. Additionally, I mean, I’d say after about a day of tuning it up, the news feed that it gives me, I mean, it’s actually pretty good. And on the

09:50 personality front, I of course ended up swapping out Jolly, the uh the fuzzy marshmallow for Flamethrower Girl. Uh and then I I had to go out and research Flamethrower Girl’s personalities from my video. And yeah, so now I’ve got a fun assistant. And overall, I mean, I have to say it is pretty efficient. I’ve been pretty much hammering Muse uh over the last four days, and I’ve only used up about 50% of my free plan. Again, that’s the free plan. Although, I’ll admit that Muse is not without its issues. For one, from like a product standpoint. I mean, I’m I’m defin I don’t think it’s meant for power users. I think this is more of a product for

10:20 soccer moms, whereas OpenAI’s dots do seem to be like I don’t know, again, really kind of hammering home like this is for serious work. And look, I don’t want to necessarily discount Muse completely. it is running spark uh which means that in terms of like model intelligence it’s somewhere around the opus level but I’ll say that some of my frustration with using m views uh does come down to the fact that I mean meta I mean there’s the privacy issues you cannot get around that I remember kranch analytica I’m sure you do as well and look I mean that might be somewhat hypocritical on my part considering that

10:52 both like openai and anthropic pretty much have free reign on my computer any major tech company is like you know one bad day away from being meta in like the the mid210s. Um but also I mean on the other side meta has a rep and it’s actually their responsibility to correct it. So it does feel a little bit sus when you know I want to give Muse sandbox access to one of the folders on my desktop only to discover that I have to give it access to my full computer. I mean that’s not really earning trust back. And just

11:22 circling back to Devday, that seemed to be an open shot that uh OpenAI I mean just had to take um with its new privacy feature. Now, they did not name Meta by name, but they’re previewing something called OpenAI private intelligence, which is a business feature where your data isn’t kept uh and the, you know, like OpenAI staff can’t see it, which I’m not exactly sure where like the deferring line here is. uh considering that dot uh works on its own cloud computer and your computer is supposed to stay separate unless you choose to connect it but at the same time uh does

11:54 that mean without the private intelligence like feature which I presume is paid um that does mean that open AAI you know can essentially train and or look at your uh your data I mean I’m not sure now the other frustration and major issue that I kept running into with Muse was its lack of connectivity although that was fixed earlier today when the Meta team announced like a whole boatload of connection that uh Muse can now well connect to. So this does bring us to a pretty interesting place where on the one hand uh you know you have open AAI with sort

12:25 of it it’s a serious dot uh which is capable of doing you know obviously like heavy lifting work and is powered by Astra but uh also costs uh $200. Whereas on the other side, you do have Muse, which is uh more built for trivial things, but clearly with integrations getting better, but is also Meta, who does have a bit of a like, you know, a privacy um rep, shall we call it? So, as I’ve said before, I think this is going to be a really interesting battleground to watch because this is essentially

12:57 like all of these companies fighting it out to become your new AI executive assistant and best friend. It is kind of funny like playing around with Muse and sort of digesting everything that we saw at Meta Connect, especially with the VR goggles and the uh uh the glasses uh that it did very much remind me of the AI film that I made a while back called Paperclip. Uh for some reason uh the YouTube algo kind of buried it. So I know I I know not everyone saw it. Um but yeah, I mean it’s linked down below if you want to check it out. Kind of kind of kind of relevant for everything

13:28 that’s going on. So, uh, yeah, if you do happen to check it out, uh, let me know in the comments there or just let me know in the comments here. As always, I thank you for watching.