AI News & Strategy Daily | Nate B Jones
Enterprise AI Pricing: Why Everything You Know Is Wrong
Ask the tutor about this video
Sign in with a verified email and ask the tutor anything about this video.
TL;DR
Research summary
▶ Pricing misses the value question
The speaker says enterprise AI deals often stall over per-seat or per-token pricing, which he views as traditional product thinking. He frames intelligence as a new economic input alongside people and capital. In a Cedar example, a hypothetical patient calls Cora about a medical bill; the agent verifies the patient, explains charges, confirms insurance, and lets the patient pay during the call. The speaker says the hospital cares that the bill is paid, not how many tokens the call used. He argues that AI agents can deliver intelligence directly to customers, which changes how companies think about value and jobs. For CFOs, metering spend does not ensure a return if the company cannot allocate intelligence well. For employees, he sees an opportunity to become managers of intelligence by directing AI toward valuable work.
▶ From AI user to AI manager
The speaker explains that per-seat and per-token purchasing puts the burden of turning AI into ROI on the buyer. Token pricing can help budget and meter model processing, but tokens do not represent a fixed amount of useful thinking; companies still need to decide what work to assign and how to judge results. He says many companies constrain spending because they have not yet learned to capture ROI at scale, while arguing that companies are under-investing and should consider ambition, employee agency, and alignment. General-purpose AI can be purchased before an organization knows which work to assign, how to change processes, or how to evaluate answers; impressive demos do not guarantee organizational readiness. He says contributors are also AI managers and raises motivation, equipment, leadership support, and mission alignment as relevant people questions.
▶ An equipment company learns where AI fits
The speaker introduces a hypothetical equipment company that helps customers keep machines running, saying the scenario closely matches stories he has seen. An engineer can use an agent to gather maintenance records and compare cases, while applying knowledge of operating conditions, data reliability, and faults that may be described similarly. The engineer must check evidence, investigate root causes, decide what to ask the AI, and know when to stop and involve a human or customer. Across enough cases, repeatable opportunities could support ongoing preventive advice and a new product, though packaging and pricing it would take work and would not happen in a week. The speaker contrasts this learning process with older paths such as developing the capability internally over a year, hiring experienced people, or working with a vendor.
▶ Forward-deployed engineering and partnerships
The speaker describes forward-deployed engineers as people who work with AI inside or alongside customer organizations to turn technology into functioning systems. He says the work can include identifying opportunities with business leaders and frontline teams, then designing, building, testing, and deploying systems connected to customer data and processes, with attention to everyday operations. He cites OpenAI and Anthropic as examples of companies using this approach; he also mentions an Anthropic joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs, working with businesses including mid-size manufacturers and regional healthcare systems. He cautions that staying in a service arrangement forever may limit learning, while pushing a product too soon may also be premature. He advocates balancing learning with frontline employees who are empowered and rewarded for applying intelligence.
▶ Outcome pricing and deeper commercial arrangements
The speaker says OpenAI took an ownership stake in Thrive Holdings and announced plans to embed research, product, and engineering teams in its companies, initially focusing on accounting and IT services. He argues that AI suppliers may bring expertise beyond model access, potentially supporting outcome pricing or joint ventures and deeper partnerships. In those arrangements, the CFO weighs potential value, what the firm retains, and what it must commit. The speaker says not every company will enter partnerships or shift to outcome pricing, which is hard to measure, but companies can experiment with structures. He emphasizes that frontline people make many AI allocation decisions and need motivation, company support, and a stake in delivering customer value.
▶ Where outcome pricing fits; token pricing remains
Returning to Sierra and Cedar, the speaker says Sierra is exploring per-outcome deals and that a completed service interaction, such as explaining and collecting payment on a bill, can in principle be priced by outcome. He does not know Cedar’s exact contract terms. He says outcome pricing fits closed-loop work such as customer service more cleanly than complex work such as redesigning a manufacturing service program. That complex work still requires employees to be equipped to deliver value with purchased AI. He expects per-token pricing to remain useful when a company knows how to apply intelligence, since usage gives room to experiment and retain gains without sharing ownership for every model use. He expects different commercial arrangements depending on companies’ AI fluency.
▶ Examples of AI returns and organizational fluency
The speaker attributes to Shopify COO Jess Hertz the view that engineers should use the best models for the circumstances, with spending monitored and adoption distinguished from impact. He says Shopify's headcount had been flat, hiring deferred for about two years, and revenue had grown 34% in that time; he presents this as an example of AI productivity contributing to business results. He also says Block discussed 25% gross-profit growth, a slimmed-down team, and AI at the heart of its business, while noting that future results remain to be seen. He argues that companies have different AI fluency and do not need identical commercial arrangements. Less fluent companies may need people changes, partnerships, deployed engineers, or software with consulting to build deep team fluency and produce results that can be confidently attributed.
▶ Career opportunity: manage AI
The speaker argues that employees can become managers of intelligence before receiving management titles, directing a scope of work beyond what they could personally do. He says less AI-fluent companies need champions who can demonstrate how to apply AI to improve the top and bottom line. In his view, frontline employees are often leading this work, and leadership can learn from them. He encourages employees to build AI fluency and pursue the role of AI manager as a career opportunity.
▶ Leadership: evaluate value and test ambitious projects
The speaker advises CFOs and CEOs to evaluate a company's ability to create value alongside ordinary AI pricing questions. He encourages leaders to ask whether the company can apply AI and generate value, and to focus on potential upside as well as cost reduction. One proposed test is to choose a revenue-generating project the company previously could not resource, assign a team to it, and attribute resulting revenue to AI if freed-up productivity made the work possible. He says organizations are still learning how to create value with intelligence and invites viewers to share their experiences with AI management and pricing in the comments.
Full transcript
[0:30] and capital. Now we have a new one. We have intelligence. Let me show you how things could look different from here. Cedar is a healthcare payments company. So far, so good. It runs an AI agent, of course it does, called Cora. A patient can call with a medical bill, even at 2:00 in the morning. Cora picks up the phone, bright and chirpy. It checks who they are, walks them through the charges, confirms insurance, and then lets them take care of their bill right there on the call. For the patient, that's nice. A confusing bill
[1:00] can get explained, and it can get handled in a call. No hold music needed. The hospital doesn't care how many tokens that call took. It cares that the bill got paid. So, in that world, we need to think differently about what intelligence brings to business and customer relationships, and we need to think about how we value it differently. And if that sounds heady, we're going to get specifically to why I think this is the key unlock for AI and jobs in 2026. Fundamentally, we are moving into a world where AI agents are directly
[1:33] bringing intelligence to create economic value, straight to the customer. And that changes how we think about what corporations do. And it changes how we think about what all of our jobs are. If you are a CFO, this is going to create big problems for you from a software perspective. Your company may know exactly what it's paying for if it charges per token, et cetera, but it may not know how to earn a return on that because it doesn't know how to allocate intelligence. And that changes what a partnership is worth to you. Now, if
[2:04] you're an individual, this creates a massive opportunity for you. You can become a manager of intelligence. And in fact, that's what everyone wants you to become from a leadership perspective because people in leadership need individuals to figure out how to harness intelligence to deliver value for the company. This is something that everybody's figuring it out together. And so, learning to take that resource and turn it into valuable work can change the scope of what you're entrusted to do. Think about a report
[2:34] that you produce regularly. You can ask an agent to finish it faster or you can ask it to investigate the root cause customer problem and start to fix it, right? That's a very simplistic answer, although I've literally seen that play out. But those kinds of mindset shifts are examples of how people can move from just being users of AI to being managers of AI. Your company has probably made intelligence available to you because, hey, you're watching this channel. You are deciding what some of that company's
[3:04] resourcing will do. You are allocating intelligence, and your decisions have real economic impact. And we have not become very good at recognizing that yet, either from the leadership point of view or as individuals. Now, why am I talking about pricing here? Why does pricing come up? I think one of the fundamental issues with purchasing intelligence as a business is that it's difficult to value. You can put it in familiar boxes as a procurement team, right? You can buy per token, you can buy per seat, and that's often the standard approach. Is it a service or is
[3:35] it a product? But then you are assuming as a company that you know how to value and turn that intelligence into something that generates real ROI. And so many companies don't. They need help turning this into ROI. And so, if you're wrestling with this challenge, what I would encourage you to do is to first step back and recognize why the current system doesn't fit well. Token pricing can fit your budget and help you manage and structure expectations for spend, and that's not a bad thing. I'm
[4:05] not saying abolish token pricing. Tokens are units that are useful to meter the information that a model processes. They have value. But, they don't actually offer a fixed amount of useful thinking. Whether that token is useful is up to all of us at the company, right? It's up to us as managers of AI. And the usual contract we get leaves that consequential assumption, that burden on the buyer. You are purchasing and you'd better
[4:36] capture enough value to justify the cost. And I think that that, at root, is why we see companies trying to figure out how to budget and constrain spend on AI. Companies have not figured out yet how to capture real ROI at scale from intelligence, and for now, they want to experiment and constrain the ceiling on their spend to avoid deep downside. And I get that. But, I got to tell you, you're under invested. You are not investing enough. Now, you have to
[5:08] figure out what to invest in, and that's a different question. That's a question of ambition. It's a question of agency. You have to think about how to empower your employees to make the right bets at scale and align them. Those are hard questions. Those are not questions you can ask Claude to solve for you. But, they're the right questions, and I think the people and companies asking those questions are the ones that will up their investment in AI and reap the most benefits. Put it this way, with general purpose intelligence, so much of the value in that intelligence needs to be
[5:38] discovered by the human using it. You can buy the capability before you've learned which work to assign to it, before you've learned how to change the surrounding processes or even before you've you learned how to tell when an answer is good. And a model can be very impressive in a demo while an organization remains very poorly equipped to use it. I see that a lot. And I think that missing know-how explains a huge part of the struggle with AI returns. A procurement process can establish that you have bought
[6:08] intelligence at a competitive price, but your investment case still depends on all of us out there who are AI managers to make it productive. Even individual contributors, we're AI managers, too. And so, the question for us is are we able to unlock that value? Do we feel motivated? Do we feel equipped? Do we feel like the company believes in us? Do we feel aligned to the mission? These are people questions, but they're real relevant in the age of AI when we're all allocating intelligence. Let's consider
[6:39] an example. Let's consider an equipment company that helps customers keep their machines running. This company is hypothetical, but it very closely matches real stories I've seen. I've talked before about how cheaper intelligence can make more extensive customer support affordable. I want to follow that investment decision through here so you get a sense of that. Who actually can build and run the service here? What should the company pay for the outcomes they're getting? Suppose you're an engineer at the company. You recognize a recurring
[7:09] failure that customers keep reporting. An agent can pull together maintenance records and compare previous cases for you. You know which operating conditions matter, which records are not reliable, and when an apparent pattern is actually two different faults that are appearing in the same language. That kind of knowledge changes how you think about this because you can dig in and ask for the right comparisons and check the evidence and discover kind of the core of the problem to help pull apart the root causes. And from there,
[7:40] you actually need to dig in with a customer to understand what to do next. At that point, sending the AI agent off to generate more explanations doesn't fix the problem. You have to manage the intelligence and make the call both when to dig in with your AI and when to stop and talk to a human. And so, ultimately, as an individual contributor engineer, you're deciding what work deserves more investigation for the company, what the machine can handle, where another person needs to step in, where you need to stop to a
[8:11] human. All of that is on you. Now, imagine that you've done this across enough cases that your team can start to see repeatable opportunities here. Customers might start to pay for ongoing advice that helps them prevent failures. The company now has relationships with those customers, knowledge about the equipment, and people who understand what a credible recommendation requires. And it is starting to turn into a product that the company can sell that wasn't there before.
[8:41] Now, someone is going to have to sort of connect that up and price it and package it and all of that. I realize that this is not as simple as ABC and it's done in a week. But, this kind of journey is exactly what I see when I see excellent AI management at work in companies. Previously, the company might spend a year discovering how to do these kinds of things itself. It might hire people who already know. It might work with a vendor. And the expected return on those paths are all measured in terms of what
[9:12] the company can execute in the old ways. And this is where I find the rise of forward-deployed engineering as a pathway so interesting. These are engineers who work with AI, who manage AI inside or closely alongside a customer's organization in order to turn technology into functioning systems. They help supply you with the ability to take intelligence you've purchased and turn it into value. OpenAI's deployment company describes its engagements in
[9:42] exactly those terms. Its engineers work with business leaders and frontline teams to identify opportunities, and then they design and they build and they test and they deploy systems connected to the customer's data and processes. The work begins by choosing where intelligence is going to apply, and it continues all the way through making sure that pays off in everyday operations. Anthropic is very similar in the language they use when they talk about forward deployed engineering and the joint venture they announced with Blackstone and Hellman & Friedman and
[10:13] Goldman Sachs. They're working with businesses like mid-size manufacturers, regional healthcare systems, and ultimately trying to actually apply AI in ways that are relevant for those companies. This is I think forward deployed engineering is a wonderful way to understand what good AI management can look like if it is placed within a structure that allows a company to harvest that value over time. If you just keep it at forward deployed engineering as a service forever and ever, you're probably not learning
[10:44] enough. If you try and push it as a product right away, you're probably not learning enough, too. There's a balance here, and part of what I'm trying to argue for is that we need really sharp individuals on the front lines talking to customers who are empowered and feel rewarded for their investment in the company, who can in turn apply that intelligence to generate value. And sometimes these kinds of opportunities are large enough to justify a relationship that goes farther, and I want to talk about that because this is the kind of difference intelligence can
[11:14] make in the economic model. OpenAI took an ownership stake in Thrive Holdings and announced that it would embed research and product and engineering teams inside Thrive's companies. The initial focus has been accounting and IT services, and we'll see where it goes from there. But, this is something I want to highlight because when we start to think about how intelligence is valued, we have to recognize that an AI supplier may have an extraordinary amount of expertise in their field
[11:45] whether or not they're a lab, right? They might not be a lab. There's other companies in the space that are doing similar work. But, the AI supplier may have the expertise to deliver value that the company needs that goes well beyond the token pricing. And so, in those worlds, you know, this can be called outcome pricing. It can be called joint venture in some cases where there's appetite for it. But, fundamentally, there's a deeper partnership involved where both sides are committed to working out together how to leverage this new economic input to unlock disproportionate value for customers.
[12:16] And so, if a partner can help make that business achievable, the CFO has a really different decision to make. It's not about the procurement team negotiating a software discount at that point. It's about thinking about how much more value could a partnership create. What would the firm retain in that scenario? What would it have to commit to get there? What is the upside and is that worth the, you know, the give and the get that everybody has to go through to get these deals done? I'm not saying everybody's going to enter into partnerships tomorrow. I'm not saying the world is all going to pivot to outcome pricing tomorrow. Outcome
[12:47] pricing is hard to measure. But, I think if we have the first new economic input in half a millennium, it probably does make sense to experiment a little bit with our structures to make sure that we are able to apply that input in ways that benefit our customers. And that goes back to us as individuals. And that's one of the things do that you'll notice I keep going after. These other traditional economic inputs are really managerial. If you are a traditional officer of the company, you can apply your team and you can apply your budget
[13:17] to get work done for the company. And kind of the employees have to go along and and do what they're going to do. But, not with AI. Not with AI. AI is something we all apply and that's why I've taken care in this video to call out examples where really it's frontline engineers, it's frontline customer facing people who are actually figuring out when to use AI and when not to use AI, and actually getting customer value across the line and ultimately creating enterprise value. That is only going to happen if they feel motivated to do
[13:47] that, if they feel like the company supports them doing that, if they feel like they are in it and they have skin in the game and it's worth going after customer value. It has never been more important to have a motivated team. A motivated team makes a big difference when it comes to applying AI effectively. Now, let's go back for a minute to Sierra. So, I started this whole conversation by talking about Sierra and Cedar and this whole idea that you might pay for delivered business outcomes. That's something Sierra's actively explor- exploring.
[14:17] That's something that they are finding deals on where they can say, "This is a per outcome basis pricing and we feel comfortable charging that, the buyer feels comfortable buying that." The Cedar example shows how that kind of completed work can actually pan out, right? A patient can get a bill explained and make a payment, and that is a full closed service interaction that can be in principle charged on outcome based terms. Now, I don't know Cedar's exact contract terms, but I think Sierra's broader proposition here is exactly that that we should start to
[14:48] think about how we measure outcomes and how we can price against them if we want to tie AI to the value that it's creating. And that will work in cases where we have those closed loops. That will work in customer service cases are a classic example. Finn has done that with customer service cases, right? Uh it will not work as cleanly when the work is more complex. If you have a complete redo of your customer's manufacturing service program, which is kind of the example I gave earlier in
[15:18] this video, that is is the same thing. That is not a single, simple, closed transaction you can outcome price, but it is something where it's worth a deeper think to make sure that your people are equipped to deliver that value with the intelligence you've chosen to purchase. I do think per token pricing is here to stay. If your company knows how to apply intelligence, buying usage gives lots of room to change those assignments, to experiment, and to keep the gains from what develops. You don't need to share ownership every time you want to use a model, no one's going to do that.
[15:49] That being said, I do think that we're going to see a variety of models here as we basically work out how to apply intelligence in companies that have differing degrees of fluency with AI. Look at Shopify, the chief operating officer Jess Hertz describes a healthy return on AI spending that she feels good about. She says engineers should be able to use the best models for the circumstances, and she monitors spending along the way. And she distinguishes adoption from impacts. Shopify's head count has been flat, right? They they sort of deferred hiring for uh about 2
[16:21] years now, and revenue has grown 34% in that time. And so from her perspective, she is getting real ROI on her AI investments because she's been able to really ramp productivity in ways that affect the top line of the business. I think that that is at least a public example of what it looks like when you move from these individual stories that I like to tell on this channel into something that affects quarterly reporting, into something that gets into a CFO's numbers. There are other stories, too. Like Block has talked about their gross profit growing 25%.
[16:52] They famously really slimmed down their team. They put AI at the heart of their business, and they have really, really leaned in on what that looks like. And so far, they've been able to show uh profit growth as a result. We will have to see how that continues to pan out. Every company will have a different story here. But ultimately, organizations are learning to make intelligence productive, and I don't see a case where every firm needs the same commercial arrangement to get there, but I do see that the commercial arrangements are different depending on
[17:22] the AI fluency in the company. Shopify and Block are both extremely AI fluent companies. I would expect them to show ROI at this point, frankly. A lot of companies I talk to are not as AI fluent, and that's the situation where you need to think about what are the people changes you need to make to ensure that you have really smart people applying AI so that you can generate that value when you put AI to work on your hardest problems. And that can look different, right? That might look like a partnership, it might look like fully
[17:52] deployed engineers, maybe it is a traditional software model with consultancy. There's lots of ways that companies are doing this, but you have got to start to get the fluency in using this new economic input deeply embedded in your team to start to get your top line results to really shift in a way that makes the ROI conversation not something you want to squint at, right? But something you can say, "Yeah, I can confidently point at that." And I don't want to lose sight of individuals in this, right? I keep circling back to this. I think this is an important moment to find our place as AI changes
[18:25] work. We can become managers of intelligence before anyone gives us a management title. And the scope of work that we can direct can easily grow beyond what we could personally do as a result. So, our judgment about what intelligence belongs where can become extremely consequential. And in my experience, especially if the company is not super fluent in AI cuz they need champions. They need people who know how to do this. And so, if you work at one of those companies and you're like, "Nate, I watch your channel. I don't know what to
[18:55] do. My company doesn't get it." You can be a manager of AI at your company, right? You can be in a position where you can show through your fluency how to apply AI in ways that really drive the top and bottom line for the company. And that's going to get attention. It's very simple. This is a new economic input. If you serve customers effectively, you're going to create more value, and that's going to accrue to the top and bottom line. And so, that's something that we've been able to do as individuals that I think is new. It's not something
[19:25] I've seen us have the power to do before, and it's something that I think is really empowering. In a lot of the cases that I see, it's actually the individuals on the front lines who are leading this charge. They see the opportunity that these tools have, and they're eagerly putting them to work. And this is me cheering you on. This is me saying, "It's absolutely the right approach, and leadership is learning from you how to think about this more strategically." And so, if you're wondering like, "Where's the career payoff here?" That's it, right there,
[19:56] right? Like, be a manager of AI. Lean into that. You're not going to regret it. Now, a word for leadership, too. If you're a CFO, if you're a CEO, if you're evaluating the next substantial AI investment, put the company's ability to create value on the table alongside the ordinary pricing conversation. Ask the hard questions about whether or not your company has the ability to leverage the AI you're purchasing and confidently generate value. I find so often we get
[20:28] into the wrong conversational track when we just think about cutting costs. We want to be in a position where we're really thinking about freeing up upside for the business because there's no ceiling on upside. You can generate a lot of top side value if you're willing to think boldly. A good example of that is let's say that you expect productivity to improve at the company if you adopt AI, then challenge that. Test it out. Pick a project that will generate revenue that you have not been able to get to, and say, "We are going
[20:59] to put a team on that project, and we would not have been able to resource that before, and now we are. And now, if we get value back, if that starts to earn revenue, we're going to attribute that to AI because we freed up productivity and we were able to do that. That kind of concrete thinking I find is actually not particularly common, and it's super important if you actually want to figure out how you were going to apply AI in a way that shows that you're really thinking through fluency and how to use this new economic input to generate value for customers. I
[21:31] know that this has felt perhaps a little bit like an economic primer. I think there's no way to get away from understanding the economics here. If we want to understand our place in it, our careers, how we as leaders need to act. If you're in leadership right now, and you're trying to figure out what to invest in with AI, or what you should even be paying for with AI, I want to tell you you are not alone. I work with a lot of teams, and I have to tell you everyone's reckoning and wrestling with this same question. We are figuring out
[22:01] how to create value together with intelligence, and it's it's a new process, right? It takes new thinking. So, this is your opportunity to sound off. Tell me how you are a manager of AI. Tell me how you are thinking as a leader about capturing the value of AI in your company. Where is it working? Does per token pricing work for you? Does outcome pricing work for you? Have you seen good examples? I'd love to hear it. Share in the comments. We'll talk soon.