Can OpenAI Really Charge $20k/Month for Agents?

Today on the AI Daily Brief, OpenAI's big agent plans involve charging up to about a quarter million dollars a year for an agent before then in the Headlines Speaking of agents, Amazon forms a new group focused on, you guessed it, Agentic AI. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show. Notes welcome Back to the AI Daily Brief Headlines Edition. All the daily AI news you need in around five minutes. Agents, agents, agents. That is the theme of 2025, as it will be in 2026 and 2027 I believe. But in any case, the latest company to jump into the agent pool is apparently Amazon. Reuters reports that the company has formed a new group internally focused on agents. According to an internal email, the group will be led by longtime AWS executives who have been involved in the AI and data teams. In the email, AWS CEO Matt Garman wrote, agentic AI has the potential to be the next multibillion dollar business for aws. We have the opportunity to help our customers innovate even faster and unlock more possibilities, and I firmly believe that AI agents are core to this next wave of innovation. The new division comes as part of a larger restructuring, with the bedrock and SageMaker AI teams being folded into the AWS compute organization separately. Amazon is not giving up the ghost on their own models According to Business Insider, Amazon will be launching a reasoning model tentatively in June under the Nova model family. The model will reportedly borrow a hybrid architecture from Anthropic, meaning that it'll be able to decide if it needs to think more and use inference time, or if that's not necessary for the particular query. Sources say that the chief focus will be to deliver a more cost efficient model, which is what it looked like Nova was trying to do when it was announced last year. Nova is offered for 75% cheaper than comparable third party models on Amazon's Bedrock. Still another goal though, is to ensure the model can rank in the top five for performance in coding and math benchmarks. I gotta say, even if it feels like they have an uphill battle, at least they're doing more than Apple. Next up, a big announcement from Perplexity. The company is teaming up with Deutsche Telekom, the fifth largest telecom in the world, to produce an AI phone. The device is intended to be a low cost handset with a new built in AI assistant called Magenta AI, and it's pretty clear from the announcement that not only does Deutsche Telekom want to get in on the AI game, but Perplexity is thinking about its future as an AI assistant/agentic assistant, even though that's not the word they used during the announcement, CEO Arvan Srinivas said Perplexity is transitioning from just being an answer machine to an action machine. It's going to start doing things for you, not just answering questions. It's going to be able to book flights, book reservations, send emails, send messages, blaze phone calls, all the sort of things that I frequently on this show rag about people not actually caring about. However, if it's totally integrated into your phone experience, maybe I'm wrong. Right now we don't have a ton more details about the phone, but this is a big score for Perplexity and will be exciting to see what they can do here in OpenAI land, the company is beginning to roll out GPT 4.5 to plus tier users. The rollout is a bit slow, but they're worried about people having a bad experience. Sam Altman writes, there's no perfect way to do this. We wanted to do it for everyone tomorrow, but that would have meant we had to launch with a very low rate limit. We think people are going to use this a lot and love it, so we think it's better to let people have real long conversations with it, but that means we have to stagger people in rather than have everyone hit it hard at the same time. Altman had previously said that GPT4.5 is the first time people have been emailing with such passion, asking us to promise to never stop offering a specific model or even replace it with an update. I think people are still figuring out exactly what 4.5 is good for. Certainly it is in a totally different class when it comes to writing, which is great because that's a key use case for me. And it sounds like I'm not the only one having an experience where even if it's not some step change better than the rest, it has some very specific uses that are really good. Lastly, today some M and A news cloud provider Core Weave has agreed to acquire developer platform weights and biases for around $1.7 billion. WNB is one of the leading providers of model training, fine tuning and deployment tools. On the model deployment side, they work with over 1400 organizations to deploy and monitor AI in production and overall the platform is used by over a million AI engineers. Coreweave is of course headed into a much anticipated IPO and in advance of that, one of the challenges that while they've seen massive top line growth 700% over the last year, reaching 1.9 billion revenue. Half of that came from Microsoft and another quarter from Nvidia. And so maybe they're looking to weights and biases as a way to diversify their customer base. Constellation Research writes that the play here is an end to end complete solution. They write. Enterprises want a turnkey cloud platform that lets them build and operate their next gen AI applications and do everything associated with them in one place. And that's what Core Weave aims to give them. Congrats to both Core Weave and weights and Biases. That's going to do it for today's AI Daily Brief headlines. Next up, the main episode. Today's episode is brought to you by Vanta. Trust isn't just earned, it's demanded. Whether you're a startup founder navigating your first audit or a seasoned security professional scaling your GRC program, proving your commitment to security has never been more critical or more complex. That's where Vanta comes in. Businesses use Vanta to establish trust by automating compliance needs across over 35 frameworks like SoC2 and ISO 27001, centralized security workflows, complete questionnaires up to 5x faster, and proactively manage vendor risk. Vanta can help you start or scale up your security program by connecting you with auditors and experts to conduct your audit and set up your security program quickly. Plus, with automation and AI throughout the platform, Vanta gives you time back so you can focus on building your company. Join over 9,000 global companies like Atlassian, Quora and Factory who use Vanta to manage risk, improve security in real time. For a limited time, this audience gets $1,000 off vanta@vanta.com NLW that's V A N T A.com NLW for $1,000 off. There is a massive shift taking place right now from using AI to help you do your work to deploying AI agents to just do your work for you. Of course, in that shift there is a ton of complication. First of all of these seemingly thousands of agents out there, which are actually ready for primetime, which can do what they promise and beyond even that, which of these agents will actually fit in my workflows? What can integrate with the way that we do business right now? These are the questions at the heart of the Super Intelligent Agent Readiness Audit. We've built a voice agent that can scale across your entire team, mapping your processes better, understanding your business, figuring out where you are with AI and agents right now in order to provide recommendations that actually fit you and your company. Our proprietary agent consulting engine and agent capabilities knowledge base will leave you with action plans, recommendations and specific follow ups that will help you make your next steps into the world of a new agentic workforce. To learn more about Super's Agent Readiness Audit, email Agentuper AI or just email me directly NLWsuperAI and let's get you set up with the most disruptive technology of our lifetimes. Welcome back to the AI Daily Brief. Today we have a story that honestly isn't even really all that big. It's a leak from a deck that says a thing that might happen in the future, but isn't for sure happening. But the amount of conversation that it has generated with wildly divergent points of view actually gives us a little bit of a preview into a very near future and serves as a jumping off point for a really interesting conversation around how agents are going to be integrated into the workforce and and how they are going to be priced. So the specific catalyst for this conversation is a report in the information that OpenAI is planning on charging up to $20,000 a month for agents. Now this all comes, it appears, from conversations and decks for investors. What's notable about that is simply the fact that OpenAI have of course an incentive to suggest the high end of what they think the pricing might be for these agents, given that those price tags are probably embedded in their rather high revenue projections. So what did the information actually see? Well, apparently OpenAI is planning on selling low end agents, which are the equivalent of what they call high income knowledge workers, for around $2,000 per month. Mid tier agents for specifically software development could cost $10,000 a month and their highest tier of agents, acting according to these sources as PhD level research assistants, could cost $20,000 a month. The information source says that OpenAI expects somewhere between a fifth and a quarter of the company's revenue to ultimately come from agent products. As you might imagine, these big eye bulging numbers definitely caught people's attention. Chubby on X writes, a corporation will only pay these prices if it receives a corresponding value from the agents and I trust OpenAI to deliver. In any case, these costs indicate that the upcoming agents, like the software engineering agent, are really so good that they can easily replace humans and are more productive in doing so, otherwise the prices would not be justifiable. And this is, I think, one really important takeaway. If OpenAI actually does bring an agent to market that is priced at anywhere near these prices, it is absolutely priced in a way that companies are going to have to view it as labor replacement, many pointed out though, as did Hugh Pham here. In case you've forgotten, most PhD students, including the brightest stars who can do way better work than any current LLMs, are not paid $20,000 a month. Saleh Kamal, Ph.D. writes, Instead of an OpenAI agent, you can hire me for the affordable price of 10,000amonth. Alexander Doria didn't really buy the framing. OpenAI should really stop with the PhD level routine. If you're selling $20,000 a month for some software model bundled agent for targeted industries, just say so. Giglio Rose points out problems with OpenAI's plan to charge so much for agents. One anthropic is far ahead in coding capability. They are also cheaper for agents as well. 2. A PhD makes a fraction of the $20,000 per month. This is just nonsensical charging. Three competition will charge a fraction I think when it comes to the anthropic encoding piece. Let's hold that aside because you have to think that OpenAI is planning on being at parody or state of the art when it comes to those things. Whether they can get there or not, who knows. But I think that this pricing is sort of predicated upon that. When it comes to the PhD making a fraction of 20k stuff, this one is hard for me to really analyze without having seen the actual materials. My guess is that OpenAI is either not really explaining accurately what they mean when they say PhD level researcher, or it's being lost in translation to the extent that they're charging $20,000 a month, which is a quarter million dollars a year for an agent. You have to think that this thing is not just a one to one replacement for a researcher, even a really high end researcher. It has to be seen as something that can do far more. In other words, this isn't 20k replacing one person, this is 20k who's replacing, I don't know, five people or something like that. I should also caveat right now that when I say replacement I'm talking about the dynamics of how people are thinking about the pricing, not advocating a that companies fire their people for agents or b even predicting that that's how it's going to play out. As I've said many times, I think that the companies that succeed in this next era are going to be those who realize that they can do way more with AI and literally invent the future and that the companies who just view it as cost savings are going to have a very short term gain followed by incredible destruction and out competition. But Still, I think it's an important heuristic to understand this pricing as based on the human labor equivalent that it could do the work of. However, one of the things that I think is going to be really difficult is to try to ground the price of these agents in terms of the value they deliver as opposed to the cost to run them. Clearly, OpenAI is thinking that they're going to be able to say, hey, this agent can do as much as your other labor, but at half to a tenth of the cost. However, as many pointed out, including Gurguli here, it's very likely that competition is going to come in and use cost as a strategic advantage. If running that agent doesn't cost anywhere near close to 20k and someone can offer it for more like $500 a month, you better believe that they're going to now, will OpenAI's agents be so much better that they're worth the premium? It's totally possible, but will they be worth a 40x premium, a 4x premium, a 2x premium, a 50% premium? That's what will remain to be seen. My base case is that competition is going to drive the cost down radically. And as I've expressed before, such as in the Doctor Strange episode, I think that the paradigm of thinking about one to one replacement for task done by human to task done by agent is going to be a very, very temporary way of thinking. When we look back when the story of agents is all written, some point out that for those companies that are going to charge less, it's actually a gift that OpenAI wants to charge so much on. VATs, writes OpenAI, launching models at 2K to 20K is going to act like free marketing for startups building vertical AI agents. Not every business can afford a 20k agent, but most can afford $500. No better time to build than this. I think this is absolutely true. I think that the market is going to fill in options for all budgets very, very quickly. I also think that because the cost of building a startup is coming down so much, you really are going to see hyperverticalization and really, really specific customization that sits on top of models that if not state of the art, are pretty damn close and either has access to specialized data or or more likely specialized knowledge in UI UX patterns that fit within the business routines or perhaps the other types of software that specific verticals are already using. Indeed, Dinos writes to the people freaking out about the OpenAI 2K 20K plan leak. You realize you can just build stuff too, right? There's no magical moat that's stopping you from creating your own agentic system that does the same thing for your domain. And I think more than ever, people are actually going to enact this. Just yesterday I heard about a banker who couldn't get any of the AI tools he wanted to use approved, so he just rolled his own really basic versions that he could use on his own computer. And the one other part of the story is that I think while OpenAI is saying right now that 20 to 25% of their revenue will be coming from agents, I wouldn't be surprised if they think, or at least have a casual inclination to think, that it might end up as even more of that. As we see the commoditization of foundation models happen in real time, where no one really has a state of the art advantage for more than a few weeks or a few months, companies like OpenAI are going to have to own some part of the customer experience. Now. OpenAI, for their part, has a pretty serious stranglehold on the consumer chatbot space right now. Obviously even Anthropic and Google haven't really been able to make a dent in ChatGPT. The biggest competitor that they've had really is Deep Seek, because Deep Seq made a much better model available for free. Anthropic, as we've talked about, is clearly trying to lock in the software engineer audience and leveraging the advantage that they do seem to have there from a technical perspective to really anchor their business in that area. I think OpenAI is going to aggressively go after high business value agents. I think we're going to see them have specific customized vertical sales agents and likely agents for a bunch of other verticals as well. I don't think in this case that that means that people who are building vertical agents should just stop. I think that it's going to be a rich competitive landscape with lots of vectors for competition, price just being one of them. And as I mentioned before, the opportunities for incredibly niche customization are going to be more viable than ever before. Still, the whole conversation points to a really interesting moment at the very dawn of this agentic era and will be interesting to see by the time these agents actually come to market what the price really is. For now. That's going to do it for today's AI Daily brief. Until next time, peace.