They Spent $150,000 on AI Tokens (And Got Nothing)

Summary

Marketing Against The Grain

Episode: They Spent $150,000 on AI Tokens (And Got Nothing)

Date: April 28, 2026
Hosts: Kipp Bodnar (HubSpot CMO), Kieran Flanagan (HubSpot SVP of Marketing)

---

Episode Overview

In this episode, Kipp and Kieran dive deep into the phenomenon of "token maxing" in the world of AI-powered businesses. They discuss the exploding spend on AI tokens, what it means for organizations, and the critical distinction between maximizing AI usage (token maxing) versus maximizing business outcomes (outcome maxing). With stories from Silicon Valley, insights from industry leaders, and actionable frameworks, the hosts explore how marketers and business leaders can make smarter decisions in an era where AI resources are enticingly accessible—but potentially wasteful without strategic intent.

---

Key Discussion Points & Insights

1. What is "Token Maxing"? (00:00-01:36)

• Definition & Current Climate:
  Kieran introduces the concept with a startling example: “A single developer spent $150,000 on tokens.” (00:00)
• Kipp expands, describing the cultural shift in tech hubs: "It's become cool to spend as much money as humanly possible on AI tokens and credits...like all the tokens, all the maxing, as much as you could possibly go." (00:59)
• Cultural References: The conversation connects "maxing" (a meme around maximizing appearance or resources) from internet culture to AI usage in tech workplaces.

---

2. The Hype and Risks of Token Maxing (01:36–03:51)

• Industry Sentiments & Benchmarks:
  Kieran quotes Nvidia’s CEO:
  "He would be kind of concerned if his average developer was not spending at least $250,000 on tokens." (01:36)
• The reality: Major enterprises, including Meta and Uber, have rapidly burned through AI budgets, often “burning through 13 times more tokens this year than they were last year.” (02:12)
• Unpredictability & Accountability:
  Kipp: "Businesses are struggling...I have to run a business to know what my costs are going to be." (02:50)
• The critical question: Does higher AI/token usage correlate with real business results?

---

3. Does Token Usage Equal Business Outcomes? (03:51–05:25)

• Case in Point: Many companies exhaust their annual AI budgets in the first half of the year.
• Model Complexity:
  Kieran: “Models are getting more sophisticated… burning through more tokens. Everyone wants the best model.” (03:54)
• Disconnect Highlighted:
  Kieran: “Does AI usage and token usage correlate to outcomes or are those two things disconnected from each other?” (03:26)
• Kipp: “If you’re investing in AI but can’t explain what changed in your business because of it… you need to map your AI usage to real business outcomes.” (04:18)

---

4. Outcome Maxing vs. Token Maxing (05:25–07:33)

• Framework by HubSpot CEO:
  Kieran: "Yamini, our CEO, had this great LinkedIn post where she said, Outcome Maxin’ better than Token Maxin’.” (04:40)
• Definitions:
  • Outcome Maxing: Using AI to tangibly grow the business (e.g., more sales, better support).
  • Token Maxing: Maximizing resource usage without direct measurement of achieved results.
• Kipp illustrates: "In sales...if I get twice as many deals this month, that’s awesome. That’s outcome maxing." (04:53)
• Critical Caution:
  Kieran: "The most dangerous person in a company today is the person who is a token maxer and bad at their craft." (06:57)

---

5. Practical Approaches to Link AI Spend to Outcomes (07:33–10:08)

• Go-to-Market Examples:
  • In sales, increased AI can be directly correlated with increased productivity per rep.
  • In customer support, ticket deflection and satisfaction scores offer measurable AI outcomes.
  • In marketing, outcomes are more nuanced and difficult to measure (e.g., speed, quality, engagement).
• Costs & Future Trends: Token prices are currently subsidized but expected to rise, requiring more strategic use in future.
• The Activity vs. Outcome Problem:
  Kipp: “One of the biggest epidemics in business has been measuring and reporting on activity instead of outcomes.” (07:33)
• Sprint System: Employing strict outcome targets and sprint-based work to align AI spend with delivered value.

---

6. Management’s Role: Structuring for Outcome Maxing (10:08–13:56)

• Creative Freedom vs. Business Discipline:
  Kipp paraphrases Tina Fey’s lesson from Lorne Michaels: “My whole job is to basically edit creative people and stop them from getting in their own way.” (10:08)
• Kipp: "We’re just letting everybody take every idea...and just do it. What you actually need is...management structure to put a little bit of constraints and get the really dumb shit token maxing out of the way." (10:29)

---

7. Simple Formula for Strategy: "AI by Outcome = Strategy" (11:07–13:56)

• Formula Introduction:
  Kieran: "AI by outcome equals strategy. If you can't state the desired outcome of AI usage, you don't have a strategy; you have token maxing." (11:07)
• Example: Reducing blog post creation time from 5 hours to 1 hour is an actionable AI outcome.
• Repeatability Test:
  Kipp: “Strategy is also a set of repeatable actions... If I’m gonna use it more than once, that’s probably interesting; if it’s disposable, think about whether it’s worth it.” (13:39)
• Host Takeaway:
  Kieran: "Outcomes are what makes you money and is how you get rich...don’t just make another company rich." (13:48)

---

Notable Quotes & Memorable Moments

• On AI Burn Rate:
  “There was a single developer who spent $150,000 on tokens. And what have I told you? That company didn't really know what the outcome of that spend was. Welcome to the era of token maxin’.”
  — Kieran (00:19)

• Industry Pressure:
  “He would be kind of concerned if his average developer was not spending at least $250,000 on tokens.”
  — Kieran, quoting Nvidia's Jensen Huang (01:36)

• On the Dangers of Token Maxing:
  "The most dangerous person in a company today is the person who is token maxer and bad at their craft.”
  — Kieran (06:57)

• Management Caution:
  “We’re basically just letting everybody take every idea...and just do it. What you actually need is…management structure...to get some of the, like, really dumb shit token maxing out of the way.”
  — Kipp (10:29)

• Strategy in a Formula:
  "AI by outcome equals strategy. If you have a team using AI and they cannot have a single sentence of what the outcome is, you do not have a strategy, you have AI token maxing."
  — Kieran (11:07)

---

Actionable Takeaways (13:56–14:05)

• Framework for Judging AI Use:
  Before investing heavily in AI or approving significant token spends, ensure every project or tool can be described with a single-outcome sentence.
• Track Repeatability and Outcomes:
  Focus on AI applications that address ongoing, repeatable challenges—not throwaway tasks.
• Management’s Responsibility:
  Leadership must set strict outcome targets and operate with sprints/project accountability.

---

Timestamps for Key Segments

• 00:00–01:36: What is token maxing? Stories from the field.
• 01:36–03:51: Real-world spending, industry expectations, and the unpredictability problem.
• 03:51–06:06: Outcome maxing vs. token maxing; how to tell the difference.
• 06:06–07:33: Measuring AI’s real impact in sales/support/marketing.
• 07:33–10:08: Changing activity-measurement mindsets; project structuring for outcomes.
• 10:08–11:07: The limits of creativity and management’s role.
• 11:07–13:56: "AI by outcome = strategy" formula and applying it to your work.

---

Closing Thoughts

Kipp and Kieran urge listeners to shift their focus from AI activity for its own sake to outcome-oriented strategies. As Kieran summarizes:
"AI by outcome equals strategy and strategy is the part that you should be good at." (13:57)

Main message: In the era of AI, real business growth comes from deliberate, outcome-focused use—not from maxing out your token consumption for the sake of it. Companies should integrate management frameworks that favor ROI and continuous improvement over blind experimentation.

Transcript

[Kieran]

A Single developer spent $150,000 on tokens.

[Kip]

So we're basically just letting everybody take every idea, whether it's good or crazy, and just do it.

[Kieran]

Meta last year said they burned through a billion tokens in a single month.

[Kip]

You have to answer a series of questions to really connect the thing you're building to the outcome.

[Kieran]

I say the most dangerous person in a company today is the person who is token maxer and bad at their craft. There was a single developer who spen a hundred and fifty thousand dollars on tokens. And what have I told you? That company didn't really know what the outcome of that spend was. Welcome to the era of token maxin. We are going to explain what it is, why you should care about it, and is this something you should think about for your team? All right, Kip, Token Maxin, I think you are, like, definitely a big culprit of token maxing and perplexity.

[Kip]

I'm a token maxer. We're not look maxers. Maxing is a thing. Right now everybody's maxing everything. You've got like, cavicular. Who's this, like, viral dude for? Look, maxing, you've got all these different Maxing and maxing has come to AI and now it's become cool to spend as much money as humanly possible on AI, tokens and credits. And I was out in San Francisco for a little while, Kieran, and literally everybody I talked to is like, I want my team spending more money, more money, more money. Like all the tokens, all the maxing, as much as you could possibly go. Is that a good thing? Is that a bad thing? How the hell should people actually, like, think about it, right?

[Kieran]

I think one of the quotes that kind of kicked us all off is the founder of Nvidia Jensen was on the all in podcast and he said he would be kind of concerned if his average developer was not spending at least $250,000 on tokens.

[Kip]

How much did you spend in tokens? And that person said $5,000. I will go ape Something else. Yeah.

[Kieran]

So for people listening, what that means is he wants to pay a developer in, like, base salary and then he kind of wants to have another $250,000 that he expends on that developer, which is through token usage. Right? Like, just burn as many tokens as you can use these AI models. Meta last year said they burned through a billion tokens in a single month. Is that good bad? We don't know. The average enterprise they're burning through 13 times more tokens this year than they were last year. And the question is, Kip, I think the question that most people have, if you're a cmo, if you're, whoever it may be, even the employee, should I be burning this many tokens? Is it just like, burn as many tokens as you can and go ham? Or is there something else more thoughtful that companies and teams and execs should be doing?

[Kip]

All the businesses I talk to that are what they're really struggling with. Kieran is like, I hear that I should be spending all this money on these tokens, but, like, I also have to run a business to know what my costs are going to be and, like, how I know what somebody's going to spend. It's super unpredictable based on what they're doing.

[Kieran]

AI token spend is really surgeon. And this is what you're saying is like, well, there's just a lot of people who are now building things, who are now creating things, who are now using AI for all parts of their work. And that's represented in this here from RAMPS Data. This article came out and it basically talked a lot about why so many AI pilots and companies were failing. But I think the disconnect people are probably trying to figure out is like, does AI usage and token usage correlate to outcomes or are those two things disconnected from each other? And I think at the moment there's a lot of questions being asked, like, does it really equate to revenue? There was a great host from Uber. I think it was a Uber CEO who talked about the fact that they had burned through their entire 2026 budget already for AI and. And every company is burning through the budget. They're not the only one.

[Kip]

Most people I talk to burned through the budget the first half of the year. Karen.

[Kieran]

Yeah, because these models are getting more sophisticated. They're burning through more tokens. Opus 4.7 is a pretty hungry model for most people. They've access to all of the models, right? You get access to Claude, you have cloud seat, you get access to all the models. You can pick whatever model you want. People who are sitting in seats, most of you all are probably listening. And most people who work in companies, they're not thinking, I shouldn't use this model or that model because the tokens are just like, I'll just use the best model. Everyone wants the best model.

[Kip]

We just dropped a free resource that tells you if your team's AI spend is actually driving results, if you're investing in AI, but can't explain what changed in your business because of it. This is exactly what you need. It has two sections, eight items and a scoring framework that maps your AI usage to real business outcomes. Get it right now scan the QR code or click the link in the description below.

[Kieran]

And so Yaminy, our CEO had this great LinkedIn post where she said ICOM Maxin better than TokenMaxin. Maybe talk through that. What is an example of me optimizing for outcome maxing?

[Kip]

Outcome maxing is like, am I getting better results and is my company growing more? And so what's happening right now in Silicon Valley, Kieran, is that you've got all this token maxing largely is talked about around software developers and then measure these software developers basically on pull requests from GitHub and like how much code are they changing and improving? And those code changes are a proxy for outcomes, but they're not really the direct outcomes. And outcome maxing is like no, no, no. I am a sales rep and I have an agent for prospecting and I get twice as many deals this month. Like that's awesome. Like that's outcome maxing. That's, that's exactly what you want. And because I get twice as many deals, I get twice as much revenue from those deals hopefully. Right. Though I will say I kind of like token maxing. Like the reason for it is that one outcome you do have to maximize is learning and how quickly you can learn. And I'd argue that all these tokens right now are subsidized by venture capitalists and late stage investors. And because of that it's like it is way cheaper to learn than it possibly will be in the future. I think a lot of people think those costs are going to come down. I don't know if that's actually true or not.

[Kieran]

So let's take go to market, right? And so there's go to market functions, teams and people. And so why I'm saying that is if I say, well, we want to have a core outcome for sales and we look across the sales department and we say the more AI usage or tokens that the sales department is using, the better their PPR is like productivity per rep. They're making much more money, they're closing more deals. So that is an example where you can correlate AI usage to overarching functionality. Result support would be ticket deflection. It gets harder in marketing actually. And that's the one thing we might want to touch on is like the outcomes are more nuanced in marketing. In sales there's like a Real binary outcome. You closed a deal, you did not close a deal. In customer support, you can look at the number of tickets you close and then you can look at the kind of quality score of how people feel about your support function. Did they have a good experience that they not have a good experience? The challenge is when you get into like the individual. So to your point is like, all right, let's just maximize for learnings. We want to burn a bunch of tokens. Tokens will never be as cheap as they are now. This is a great time to learn because they're being subsidized by all of the VCs. All of these companies are running negative margins and at some point they'll have to improve those margins and the cost of a token will go up. So you'll have to be more deliberate about how you use the AI assistance. But I say the most dangerous person in a company today is the person who is like token maxer and bad at their craft.

[Kip]

So as you were talking, like, I came up with two things. One is, I think there's two things you can do to create an environment in which token maxing and outcome maxing are kind of the same thing. And the first thing that I think you would do, Kieran, is that over the last 10 years, I would say one of the biggest epidemics in business has been measuring and reporting on activity instead of outcomes. Like, I do these things whether those things are actually having an impact or not. And so the first thing that you have to do in like a token maxing world, I think is set up your team to have very strict outcome targets and to have like kind of a operating model. We use like, we're using like a Sprint system and everything to basically do discrete tasks to try to achieve those outcomes and really report against those. And I think if you do that, that's, that's one thing. The other thing is it doesn't make a lot of sense to like spend all these tokens to rebuild a web page to change the color of a button if I'm going to build a completely different styled product page and do something radically different. And the fact that I can do that faster and cheaper and I can actually run a test to see if that's going to get me a much better outcome. And I think that's a good thing. What, do you agree with these or not?

[Kieran]

Yeah. So I think you're saying, so like, if you're a cmo, for example, the best way to do this is to have a quarterly set of projects And I you want to derive from AI usage. And so you're saying, hey, like one of the things we want to do is we want to use AI across the content team and then we're going to look at the quality and speed to create content. Or the really thing we want to do is like integrate AI across the social media team to reduce our agency spend and actually improve our engagement of social media. So like there's some sort of outcome you have to derive and say, well now I can say based upon this AI usage we're actually seeing like real outcomes. There should be some way to correlate that. At some point you're going to have to say what model should I use? And I think that's the hard part for folks is like this is a simplistic task. I should use a really cheap model. I think companies will start to integrate open source models and fine tune them for their own companies and then you'll actually have tasks mapped to different models. But at the moment the user is being asked to decide the model. And I would say that 99% of all people working in companies do not think about that at all. So what's really happened is everyone really kind of fucking hates their job. They wish they could just like build shit and AI allows them to build shit. And so now I used to have to go in and do this boring thing where I change the color of the button. Now I can use AI to do it in a much more creative way or rebuild the entire page so everything looks like an AI problem to solve.

[Kip]

Here's my, my explanation of the problem. And a long time ago I read Tina Fey's book Bossypants and she has like a section there in the middle where it talks about Lorne Michaels who has run Saturday Night Live for a long, long time. And he talks about his whole job is to basically like edit creative people and stop them from getting in their own way and like kind of suppress some of that craziness. And that's kind of what's happening with Tokamaxing is we're basically just letting everybody take every idea that's whether it's good or crazy and just do it. And what you actually need is like a management structure and a team structure to put a little bit of constraints and get some of the like really dumb shit token maxing out of the way. Because there is just too much creativity, too much decision making and too much just like also just like add of like, oh well, I can build this thing now. So I'm just Going to go and just spend 30 minutes building this thing, whether I should or not.

[Kieran]

We could give people a really simplistic formula here, which is like AI by outcome equals strategy. And so what do we mean by that? We mean that if you have a team using AI and they cannot have a single sentence of what the outcome of that AI usage is, you don't actually have a strategy, you have AI token maxing, you just have usage for the sake of usage. And maybe you would debate that. That's learning and they're just going to learn. But an example would be, right, the content team are using AI and so we would say, okay, you have all access to Claude or ChatGPT or Gemini. What is like one simplistic outcome this month? And they would say, oh, on the blog, we're going to try to reduce the time it takes to create a blog post today. It takes like five hours and then we want to get it down to like an hour. But because you actually have a single sentence that explains the outcome, now you have a strategy, now you have AI usage that's going to like shorten the time to produce content. If you don't have AI by outcome strategy and you're missing that single sentence, you have token maxing, you do not have outcome maxing. And that I think is a good force and function for people who are even using the models themselves. What is like one thing that I'm going to do this week that I'll get better at? So my AI usage, aside from all of the things we're kind of doing In HubSpot, my AI usage outside of HubSpot is to build a second brain. And I could do a tutorial at some point, but at least I know what my outcome is because if I build a second brain, it's going to drastically reduce the time it takes me to get back to people and actually drastically improve the strategic decisions I make of want to spend my time on. So there's like AI by outcomes come equals a clear strategy for me.

[Kip]

Right.

[Kieran]

It's strategically important to me.

[Kip]

I really like that framework. I think everybody watching the show today should take advantage of it. The one clarifying thing I would say that you just gave in your example that I really liked. It's like you have to answer a couple of a series of questions to really connect the thing you're building to the outcome. Oh, I'm gonna. I wanna build a second brain. I wanna build a second brain because it's gonna save me time that's gonna. And allow me to Respond faster. Because I can respond faster. I'm going to be able to ship decisions faster and get results faster and that's the outcome I'm getting. Right. I think sometimes people get stuck on like, hey, I just want to build this thing without fully connecting it to the outcome that they're going to get from it existing. And also like Kieran, strategy is also a set of repeatable actions and so much AI stuff is just like one off disposable shit. Like the other filter I would give is like, I'm about to build this thing that I think is going to get a good outcome. Am I going to use it more than once?

[Kieran]

Yeah, yeah.

[Kip]

And if I'm going to use it more than once, like, oh, that's probably an interesting. If this is really like a completely disposable thing and it's going to take me more than a couple minutes, like I probably shouldn't do it.

[Kieran]

Outcomes are what makes you money and is how you get rich and you don't want to just do one side of it because you're just making another company rich.

[Kip]

Right.

[Kieran]

And I think that is the part that we would leave you with is like the AI buy outcome equals strategy and strategy is the part that you should be good at.

[Kip]

You are going to be good at and hold the people you work with to equally being good at it so that you're spending your money wisely and for things that are actually going to grow your business. I love those takeaways. Kieran, I think you nailed the close there. Please hit like, please hit subscribe and we'll see you very soon on Marketing against the Green. I want to tell you about a podcast I love. It's called Nudge. It's hosted by Phil Agnew. It's brought to you by the HubSpot Podcast network, the audio destination for business professionals and it's the UK's fastest growing business podcast. What I love about it is that the Nudge listeners love no fluff, no BS evidence based marketing tactics they get in each episode. You're going to want to listen because this is like an MBA's worth of insight in every single podcast. And entrepreneurs, you're going to love the show because it's filled with repeatable, proven studies, not hearsay, not one off success stories marketers, you're going to love it because it discusses the psychology behind great marketing and what marketers are getting wrong. Listen to the Nudge wherever you get your podcasts, this data is wrong every freaking time. Have you heard of HubSpot HubSpot is a CRM platform where everything is fully integrated. Whoa. I can see the client's whole history. Calls, support tickets, emails. And here's a test from three days ago.

[Kieran]

I totally missed HubSpot Grow better.