How to Price AI Agents (And Why It Matters)

0:00Today on the AI Daily Brief, how should we be pricing agents? Before that in the headlines, a new EO on AI education. 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.

0:23Welcome back to the AI Daily Brief Headlines Edition, all the daily AI news you need in around five minutes. We kick off today with something of an unexpected one. President Trump has signed an executive order to boost AI education and workforce training. A White House fact sheet said,

1:01leaders in AI technology also requires investing in our educators, providing them with the tools and knowledge to both train students about AI and utilize the technology in the classroom. Now, in terms of specifics, the executive order establishes a task force on AI education. Led by the director of the Office of Science and Technology Policy, the committee includes the secretaries for agriculture, labor, energy, and education alongside numerous White House advisors. The task force was instructed to establish a prize called the Presidential artificial intelligence challenge, which will, quote, encourage and highlight student and educator achievements in AI, promote wide geographic adoption of technology advancement, and foster collaboration to address national challenges with AI solutions.

1:38Alongside, the task force is instructed to develop online resources for K-12 students in foundational AI literacy via public-private partnerships. The order also contemplates the use of various federal grants to advance the goal of AI education. Teacher training is addressed with the Secretary of Education instructed to prioritize the use of AI in discretionary grant programs for teacher training, and the Department of Labor is instructed to direct funding towards AI apprenticeships and certification programs designed to provide high school students with workforce credentials for AI careers. As with most EOs, the details are left to government officials to execute.

2:10However, White House Staff Secretary Will Scharf gave the elevator pitch, stating, The basic idea of this executive order is to ensure that we properly train the workforce of the future by ensuring that schoolchildren and young Americans are adequately trained in AI tools. On signing the order, President Trump said, this is a big deal. AI is where it seems to be at. We have trillions of dollars being invested in AI. AI is the way to the future. Very smart people are investing in it heavily. Now, part of the impetus for moving on AI education is likely pressure from China. Numerous provinces in China have introduced mandatory AI education at all gray levels, K through 12.

2:42No doubt the rest of China will quickly follow after the central government recently placed AI at the center of a comprehensive education overhaul. The education ministry said that promoting AI will help, quote, cultivate the basic abilities of teachers and students and shape the core competitiveness of innovative talents. Moving over back into business world, OpenAI is now forecasting$125 billion in revenue by 2029. According to investor documents viewed by the information, OpenAI sees this revenue boost coming as revenue from agents and other products overtake those from ChatGPT. These numbers would put OpenAI firmly in the top tier of tech companies.

3:16NVIDIA recorded$130 billion in revenue last year, while Meta recorded$160. OpenAI's 2029 forecast isn't anywhere near terminal growth, with the company predicting they'll be able to achieve 40 % growth to hit$174 billion in revenue for 2030. In one sense, the company is extrapolating from very little information. OpenAI achieved$3.7 billion in revenue last year, which was almost 300 % year-on-year growth. Predicting what the next 30-fold increase in revenue looks like is speculative to say the least. But the components of this stratospheric growth say a lot about where the company sees the AI industry going in the medium term.

3:50The first point that comes from OpenAI's numbers is that the level of hypergrowth is actually a necessity for the company. They expect to burn$46 billion in cash over the coming four years, and that 2029 figure of$125 billion is actually what they need to hit profitability according to their own estimates. Part of how they get there is through cost moderating. The company is predicting that inference costs will come way down over the rest of this decade. Last year, gross profit as a percentage of revenue was only 40%. OpenAI is modeling that ratio increasing to 70 % by 2029. Interestingly, that's still lower than the average gross margin of cloud software companies, which is currently around 74%.

4:25OpenAI expects another revenue boost to come from new product lines. At the moment, revenue generation is simply two buckets, consumer chat GPT subscriptions and API usage from developers. The company sees packaged agents like Operator, though, being a big seller as they get more performant. They see agent revenue hitting$29 billion, up from$2 billion this year. Simon Smith from ClickHealth wrote,

4:57Now still, the 2029 projections view ChatGPT subscriptions as a major component of revenue, representing$50 billion in sales, which seems to imply either paid user numbers in the billions or the ability to charge premium prices and push large-scale enterprise deals. The figure is separate from API usage, which OpenAI sees growing to$22 billion by 2029, once again up from$2 billion today. Underpinning all of the numbers is massive expectations of growth in users by the end of the decade. OpenAI expects to hit 3 billion monthly active users, 2 billion weekly active users, and 900 million daily active users by 2030.

5:33They're currently at 500 million weekly active users, up 60 % since December. For some comparison, Gmail has somewhere in the range of 1.5 to 1.8 billion monthly active users, and Facebook claims 3 billion monthly active users. Speaking of OpenAI, the company's ImageGen API is now available. Developers can now integrate OpenAI's image generation model into their own products. The feature has been one of OpenAI's most successful rollouts ever, with the Ghibli trend powering 700 million uses in its first week of availability. Pricing is set to$40 per million output tokens, with OpenAI saying that this translates to around$0.02 per low-quality image generation, $0.07 on medium settings, and$0.19 for high-quality images.

6:14And while indie developers are getting their first opportunity to make use of the model, OpenAI says that the range of corporate partnerships is already extensive. Adobe, Airtable, Wix, Instacart, GoDaddy, Canva, and Figma are already using the mode or experimenting with its integration into their apps. Lastly today on OpenAI, and lastly in general in the headlines, details have been emerging about OpenAI's open-source model throughout the week. The company is targeting an early summer release and aims to produce a reasoning model that tops benchmarks. TechCrunch sources say that Aidan Clark, OpenAI's VP of research, is leading the project, which is still in its early stages.

6:47They added that the plan is to release the model with few limits on commercial use, contrasting with open models from Google and Meta that have prohibitions above a certain scale. The goal seems to be to compete with DeepSeq and other open-source models out of China, which have relatively few restrictions on how they're used. Sources say the model will be text-only, operating on high-end consumer hardware, and potentially have the ability to turn reasoning on and off. Later in the week, further reporting added that the model could include unique architecture touches to achieve those lofty benchmarking goals.

7:15OpenAI leadership are discussing plans to allow the open model to connect to the company's closed models in order to quote-unquote hand off more difficult queries. The feature was reportedly suggested during one of the recent developer forums the company is conducting to source outside input on the model, and the idea is apparently gaining traction within the company. By all accounts, the model is still in the planning phases and training hasn't begun, but there is a lot of excitement about this one and we will have to keep an eye on it. For now, that is going to do it for today's AI Daily Brief Headlines edition.

7:41Next up, the main episode. Today's episode is brought to you by Vanta. Vanta is a trust management platform that helps businesses automate security and compliance, enabling them to demonstrate strong security practices and scale. In today's business landscape, businesses can't just claim security, they have to prove it. Achieving compliance with a framework like SOC 2, ISO 27001, HIPAA, GDPR, and more is how businesses can demonstrate strong security practices. And we see how much this matters every time we connect enterprises with agent services providers at Superintelligent. Many of these compliance frameworks are simply not negotiable for enterprises.

8:18The problem is that navigating security and compliance is time-consuming and complicated. It can take months of work and use up valuable time and resources. Vanta makes it easy and faster by automating compliance across 35-plus frameworks. It gets you audit-ready in weeks instead of months and saves you up to 85 % of associated costs. In fact, a recent IDC white paper found that Vanta customers achieve$535 ,000 per year in benefits, and the platform pays for itself in just three months. The proof is in the numbers. More than 10 ,000 global companies trust Vanta, including Atlassian, Quora, and more.

8:49For a limited time, listeners get$1 ,000 off at vanta.com slash nlw. That's V-A-N-T-A dot com slash NLW for$1 ,000 off. Today's episode is brought to you by KPMG. In today's fiercely competitive market, unlocking AI's potential could help give you a competitive edge, foster growth, and drive new value. But here's the key. You don't need an AI strategy. You need to embed AI into your overall business strategy to truly power it up. KPMG can show you how to integrate AI and AI agents into your business strategy in a way that truly works and is built on trusted AI principles and platforms. Check out real stories from KPMG to hear how AI is driving success with its clients at www.kpmg.us slash AI.

9:36Again, that's www.kpmg.us slash AI. Today's episode is brought to you by Superintelligent, and I am very excited today to tell you about our consultant partner program. The new Superintelligent is a platform that helps enterprises figure out which agents to adopt, and then with our marketplace, go and find the partners that can help them actually build, buy, customize, and deploy those agents. At the key of that experience is what we call our agent readiness audits. We deploy a set of voice agents which can interview people across your team to uncover where agents are going to be most effective in driving real business value.

10:10From there, we make a set of recommendations which can turn into RFPs on the marketplace or other sort of change management activities that help get you ready for the new agent-powered economy. We are finding a ton of success right now with consultants bringing the agent readiness audits to their client as a way to help them move down the funnel towards agent deployments with the consultant playing the role of helping their client hone in on the right opportunities based on what we've recommended and helping manage the partner selection process. Basically, the audits are dramatically reducing the time to discovery for our consulting partners and that's something we're really excited to see.

10:44If you run a firm and have clients who might be a good fit for the agent readiness audit, reach out to agent at bsuper.ai with consultant in the title, and we'll get right back to you with more on the consultant partner program. Again, that's agent at bsuper.ai, and put the word consultant in the subject line. Welcome back to the AI Daily Brief. Today, we are talking about agent pricing. And while this may seem like it's just insider baseball or only relevant for startups that are trying to figure things out, I actually think that it is much more important than that. The agent pricing conversation is not just about how startups and tech companies are thinking about things, but also implicates how enterprises are imagining and considering agents.

11:24Are agents' software to be procured in the same way that SaaS was before? Are they digital employees to be hired? How does the answer to those two questions impact how they should be paying for them? It is absolutely the case that agent pricing will shape and have a deterministic impact on agent and AI business models, which will have impact on things like company design, all of which is to say this is actually an important conversation. Now, the specific context that prompts us to have this conversation today is that Windsurf has sort of kicked off a price war on AI coding assistants. Their standard pro tier is now priced at$15 per month, with an allowance for 500 prompts before requiring a top-up.

12:03They're also getting rid of their Flow Action credit system, which charges for tool calls within an agentic workflow. The company says this means you can only pay per user prompt. Tech and enterprise subscriptions also saw a price drop to$30 per month and$60 per month, respectively. CEO Rob Hu made it clear that this was a direct shot at Cursor. Although he didn't mention the rival company by name, his announcement post stated, With today's pricing update, Windsurf now has by far the most affordable pricing structure of all AI coding tools on the market. And in a very pointed comparison, he added, Compare that to any of the other tools where$20 per month gets you less prompts, confusing requests, and hidden chargers for tool calls or, the absolute worst case scenario, paying per token.

12:43And not charging for tool calls is a big deal as agentic coding gets more complex. WindSurf is charging a flat rate while complex cursor prompts can easily balloon to cost a dollar or two each. CEO Rob here is very much hustling. He spent hours this week replying to potential customers, commenting on how they want to see the service structured in price. Of course, this all comes in the background of a rumored OpenAI acquisition. The deal isn't finalized, but according to well-sourced reporting, OpenAI is looking at a$3 billion price tag to buy the company. Alongside the price drops, Windsurf announced users will get another week of free and unlimited usage of the newly released GPT-4.1 and O4 mini models.

13:20It is a question, I think, about whether Windsurf remains profitable at these price levels or if they're subsidizing users to gain market share. In their announcement post, the company said that their infrastructure engineers have been able to optimize GPU usage and that they are delivering on their promise to, quote, pass savings back to our end users. Obviously, I'm now watching closely to see if Cursor fires back to commence a full-scale pricing war in one of the fastest-growing verticals in tech history. So as I said, that is the specific genesis of the conversation today. But this agent pricing conversation is happening more broadly.

13:51Aaron Levy from Box recently wrote, Nailing AI agent pricing is a super important topic for AI companies right now. There are two dynamics at play in AI pricing land. The first dynamic is that the models themselves are getting cheaper and cheaper to run. but the other trend is that the use cases for customers are requiring more and more inference. We've seen examples of deep research using up to 100x the compute of a standard query before. AI coding agents similarly can consume enormous resources depending on the task. So even as the inference gets cheaper per token, the total inference goes up dramatically.

14:23A normal thing to do with resource pricing is to effectively shift the cost of the resource onto the customer. It's a clean model, but there are many areas where a key customer use case may be technically possible but unaffordable today, even though they will become affordable tomorrow. So, do you wait to solve the problem when it's economically practical to scale, or lean in now and bet on the cost improving? The answer probably would be different in any other technology category in history. But the implications of AI model efficiency improvements is that software companies can afford to price AI in a way that anticipates the cost curve over time.

14:53This allows you to unlock use cases today that may be otherwise less economically attractive, but where you know they soon will be. It's definitely a bet, but one that increasingly seems like it will pay off. This is all thanks to the constant AI breakthroughs coming from the Frontier Labs as well as OpenWeight's model providers. And this doesn't appear to be slowing down anytime soon. So Aaron here in this particular case is not discussing heuristics or models or frameworks for pricing, but instead just the changes in cost of goods sold, and specifically the countervailing forces driving it in two directions.

15:23Cheaper to run models, but more inference-required use cases. At the beginning of April, Manny Badina, the founder of Paid, wrote a post on the Growth on Hinge blog called A New Framework for AI Agent Pricing. The analysis comes after looking at dozens and dozens of different AI agent startups, and Manny divided their pricing models into four different quadrants. They are per workflow, where you pay per completed workflow, per agent outcome, where you pay per completed objective. So both of these are outcome-based, one based on successfully completing a workflow and one based on successfully completing an objective, however many workflows it took.

15:57And then another category is what Manny calls activity-based. So that's, for example, a fixed monthly fee per agent. This is sort of a replacement model for FTEs. Or then a pay-for-usage, a per-action agent or a consumption model. As Manny points out, these different models are good for different types of companies. The FTE replacement model, a price per agent, he says is best for AI agents handling broad responsibilities or entire job functions with consistent predictable workloads. He points out that the advantage is that you get to draw from the headcount budget or labor budget, which is at least 10x larger than the software budget, but that in many cases, this leaves you subject to other companies that just charge less.

16:35The consumption model or price per agent action, he suggests is best for agents performing very discrete tasks with unpredictable frequency or volume. The price per agent workflow, aka the process automation model, he says is best for agents that execute multi-step processes with clear intermediate deliverables. and the price per agent outcome, aka the results-based model, he says is best for AI applications with predictable performance and clearly defined success metrics in markets that already expect it. Now, I want to go back to his FTE replacement comparison to a post from last year from Y Combinator called Vertical AI Agents Could Be 10X Bigger Than SaaS.

17:09This was actually on their podcast. And basically the argument here is that if agents actually are replacing big swaths of human labor, then they're competing for labor budgets, not software budgets, with labor budgets being radically higher than software budgets. The interesting tension that we're already seeing over here at Superintelligent as companies get introduced to agents is whether they're going to be priced in reference to the equivalent human labor or in reference to the cost of goods sold with some amount of margin. I think the agent companies in general are trying to keep the comparison to the comparative human labor.

17:42And that makes sense because those budgets, again, are much larger. The challenge is a couple pieces. First of all, if you're trying to say, my agent does the work of a junior developer and a junior developer would cost$100 ,000, but my agent's only going to cost$40 ,000, it's almost for sure that someone who can provision that agent for much cheaper is going to say, well, screw that. It doesn't matter that the human equivalent was$100 ,000. The thing only costs me a couple thousand dollars to run, so for$5 ,000, you can have it. Basically, there is likely competitive pressure which pushes prices down.

18:11The other thing is that some of the use cases that agents will be used for won't be able to be priced compared to their human labor because a priori the cost of human labor would have been so high previously that no one would have ever considered doing the thing. I have a very specific example from our agent readiness audits. As part of the audits, we do voice agent interviews, which can be across a handful of leaders or it could be across every single person inside a department. Right now we're live with a big pharma company with all 200 members of a specific department of theirs. And the problem if we had tried to price that against what it would have cost for McKinsey to interview all 200 people in that department is that we know a priori that the cost of McKinsey interviewing 200 people like that would be so astronomical that they never would even consider that.

18:55So we can't sit there and say, hey, if you ask McKinsey to do this, it would be $300 ,000 and we only want to charge you$100 ,000 because they'd say, yeah, but we never would even consider paying that because it's so out of range as to be irrelevant for our actual planning. And so we're left, in our circumstance, having to price it closer to the reference point of the cost of goods sold because what we're doing is fundamentally opening up new opportunities that simply weren't possible before agents. Now, one really interesting countervailing note comes from Signal. They wrote, was talking to someone else deep in AI and we started riffing on how agent pricing could spiral into totally uncharted territory.

19:30Like if someone builds a quote-unquote perfect agent for legal diligence for biotech, and is protected by real moats, data evaluation frameworks, workflow lock-in and trust, they could charge above human rates and still be the obvious choice. 24-7, few errors, infinite scale. Kind of ridiculous that the floor for AI labor is maybe heading towards zero, but the ceiling could possibly go infinite at least in the near term. Now, this is one of those things that really requires a lot of imagination around what the moats would be that would even justify this. But I think a relevant takeaway, no matter what, is that agents aren't just going to be competing on cost.

20:05Their permanent and perpetual availability, their massive scalability, these are things that make them better than the human equivalent, not just a better choice because they're cheaper. Again, to round up with the same example of the voice agent interviews we use for the agent readiness audits, if you wake up with insomnia at 1.30 a.m., you're welcome to do your interview then. You don't have to schedule it with us. You can do it at your own convenience. And that is a massively better experience than if we were doing traditional voice interviews with our analysts. Anyway, all of this I think is interesting, and not just in a theoretical way, but in a way that will shape the future of the AI and agent space.

20:38For now though, that's going to do it for today's AI Daily Brief. Appreciate you listening as always, and until next time, peace.