Shownotes
BCG's Matt Kropp explains why productive AI use, not just access, is becoming the biggest lever for growth. Companies using AI purposefully are growing three times faster than those that aren't. So why are many still stuck? The winners aren't substituting people with AI — they're augmenting them and arming them with AI tools, building an advantage that compounds over time.
You’ll Learn:
Why “token-based competition” is emerging as the next growth advantage
What Matt learned from building — and shutting down — an all-AI company
The real reasons employees still aren't adopting AI at scale
Learn More:
The Era of Token-Based Competition Is Here. Is Your AI Strategy Ready?: https://on.bcg.com/4ARAa79
BCG’s Latest Thinking on AI: https://on.bcg.com/4hZqpw1
Matt’s Substack: https://mattkropp.substack.com/
AI Costs a Fortune. Mismanaging AI Will Cost You More: https://lnk.to/so-what-BCG-Goydan-Lukic-AI-Cost12
Chapters
0:00 Equal AI Access, Equal Results?
1:00 What Is Token-Based Competition?
1:39 Token-Based Competition vs. Token-Maxing
3:50 Inside a Zero-Human Company
5:35 Lessons from An All-Agent Company
10:03 How To Win in the Token Economy
11:06 Is AI a Substitute for Labor?
12:21 Token Use and Company Performance
14:38 What Is Return on Intelligence?
15:06 What ‘Intelligence’ Really Means
15:41 Getting Your Team On Board
17:00 Why AI Adoption Still Lags
Meet the Expert
Name: Matt Kropp, CTO, BCG X
Bio: https://on.bcg.com/4s2FuRb
LinkedIn: https://www.linkedin.com/in/matt-kropp/
Follow BCG
Website: https://www.bcg.com/
LinkedIn: https://www.linkedin.com/company/boston-consulting-group
This podcast uses the following third-party services for analysis:
Podtrac - https://analytics.podtrac.com/privacy-policy-gdrp
Transcripts
- We have ubiquitous intelligence.
Speaker:Everybody has access to the same intelligence.
Speaker:That sounds like a recipe
Speaker:for every company becoming the same.
Speaker:The reality is, today,
Speaker:most businesses are not adopting AI effectively.
Speaker:And so there is actually a real opportunity today,
Speaker:and this is if I were to give you, you know,
Speaker:to any CEOs listening to this podcast,
Speaker:a prescription on what to do with this,
Speaker:it is move fast.
Speaker:There is a real advantage to be gained by those companies
Speaker:that are first and early to bring AI intelligence
Speaker:into as much of their work as they can.
Speaker:- Welcome to "The So What from BCG," the podcast
Speaker:that explores the big ideas shaping business,
Speaker:the economy, and society.
Speaker:I'm Georgie Frost.
Speaker:What happens when AI makes intelligence abundant?
Speaker:Access alone may no longer give you the edge.
Speaker:The advantage could instead lie in
Speaker:how effectively you turn that intelligence into value.
Speaker:Welcome to the era of token-based competition.
Speaker:Joining me to discuss this is Matt Kropp, CTO of BCG X.
Speaker:Matt, you and your colleagues came up with this concept
Speaker:of token-based competition.
Speaker:Tell me, what is it? Where did it come from?
Speaker:- We wanted to look at what's the correlation
Speaker:between use of AI and success and growth.
Speaker:And so what we found was the firms that use more tokens,
Speaker:that is--and we should talk about what tokens are--
Speaker:but use more AI in their work
Speaker:actually grew faster than the firms that use less.
Speaker:So we think that the future of competition will be
Speaker:maximizing the productive use of tokens.
Speaker:That is token-based competition.
Speaker:- On a previous episode, Matt, of "The So What"
Speaker:with Paul Goydan and Vlad Lukic,
Speaker:we talked about token maxing.
Speaker:This is the companies consuming huge numbers
Speaker:of tokens creating little value.
Speaker:How do you reconcile that with the idea
Speaker:that greater productive token use could be
Speaker:a competitive advantage?
Speaker:- So I always get this question
Speaker:because I always say,
Speaker:"Token-based competition will be
Speaker:the maximizing of productive use of tokens."
Speaker:And productive is the key phrase there.
Speaker:So tokens are the language of large language models.
Speaker:Think about it as a word, and that's how they process.
Speaker:So token maxing--this concept came out of,
Speaker:there were some companies who made
Speaker:their employees' performance reviews partly dependent
Speaker:upon how many tokens they used.
Speaker:And so guess what happened.
Speaker:People started to create little scripts
Speaker:that would just run up the bill of tokens
Speaker:so that they could go up on the leaderboard
Speaker:and they could get credit for using lots of tokens
Speaker:and then they go and keep doing things the way
Speaker:that they had always done them.
Speaker:Obviously, that's not a good thing.
Speaker:That is not productive use of tokens.
Speaker:So I'm not advocating unlimited use of tokens.
Speaker:I'm saying, when you use them productively,
Speaker:when you use them for positive ROI activity,
Speaker:you should maximize their use.
Speaker:And if we go back to this idea of token-based competition,
Speaker:the notion is not that we put AI
Speaker:replacing all of the people.
Speaker:The notion is that I need the people in the company
Speaker:to direct where the AI is used.
Speaker:And so now I as the employee, I'm saying,
Speaker:how do I use this technology to do some part of my job?
Speaker:I'm applying the judgment. What should I do with it?
Speaker:I'm talking to it, in a sense, to understand what it can
Speaker:help me with, what knowledge it can bring to my work.
Speaker:I'm tasking it with doing toil,
Speaker:taking the repetitive work that isn't low value-add
Speaker:and outsourcing it to the AI.
Speaker:- Matt, you actually did an experiment into this.
Speaker:Tell me all about that.
Speaker:- I had this idea because we're doing so much work
Speaker:around AI agents and work,
Speaker:and I wanted to see was it possible
Speaker:to create a zero-human company, 100% AI company.
Speaker:So I set up, I set up a bunch of agents.
Speaker:There was a CEO agent.
Speaker:There was a product manager, a marketing agent,
Speaker:a sales agent, a customer support agent,
Speaker:and I set some rules for myself.
Speaker:Essentially, I would not make any decisions.
Speaker:The agents would make all the decisions.
Speaker:I was just there basically
Speaker:to sign the check and to be the human.
Speaker:And I let them come up with what the company idea was.
Speaker:They told me how to incorporate,
Speaker:so we have a Wyoming LLC
Speaker:that is incorporated called Vesica Labs.
Speaker:They came up with the name.
Speaker:They came up with the definition
Speaker:of what the product would be.
Speaker:They built the product. They launched the product.
Speaker:All of it was done by the agents.
Speaker:I didn't make any decisions.
Speaker:I let them make all the decisions.
Speaker:So in some ways, it was incredible
Speaker:what you can get done
Speaker:with me and a bunch of agents and no other humans.
Speaker:- You actually fired the CEO, didn't you?
Speaker:- I did, and I shut the whole thing down
Speaker:because it didn't actually work.
Speaker:And so the lesson was
Speaker:the agents are amazing at coming up with ideas.
Speaker:They will come up with ideas forever,
Speaker:but the ideas are not necessarily good ideas.
Speaker:They are amazing at building things,
Speaker:but the building would get to 80%
Speaker:and then you would look at what it created,
Speaker:and it worked.
Speaker:But as a human, it was not very useful.
Speaker:- Give me an example.
Speaker:How did you know when you looked at something,
Speaker:oh, this just is never going to work?
Speaker:Because it sounds good.
Speaker:It always sounds good, doesn't it?
Speaker:It's just never going to work.
Speaker:- Well, it's about human taste and judgment.
Speaker:You know, there was a,
Speaker:they had built this platform.
Speaker:The business that they came up with was to create a platform
Speaker:for people to create zero-human companies.
Speaker:So it was a zero-human company for zero-human companies.
Speaker:And it was a full website.
Speaker:You could log in.
Speaker:You could put in a credit card to subscribe.
Speaker:It had a well-designed layout,
Speaker:but as a human then coming in to use it,
Speaker:it was confusing.
Speaker:I wasn't sure what it was for.
Speaker:I wasn't sure exactly what benefit it would give me.
Speaker:And so it missed some element of
Speaker:what a human product manager,
Speaker:what a human designer would bring in terms of
Speaker:how it should actually feel
Speaker:and how it should actually work.
Speaker:So my conclusion was incredibly powerful.
Speaker:I'm a big AI optimist, obviously.
Speaker:I absolutely believe that in the hands of humans,
Speaker:you can do amazing things with these tools.
Speaker:And so they're very good
Speaker:at providing expertise, knowledge
Speaker:that has been encoded into those models.
Speaker:They're very good at coming up
Speaker:with plans based on some direction.
Speaker:They're very good at executing against plans,
Speaker:especially for software engineering.
Speaker:If you give them a task, they will generate excellent code
Speaker:that's fully tested, that fully works.
Speaker:What they're not good at is understanding what to do.
Speaker:But it was very clear that you need the human
Speaker:to be making the strategy calls
Speaker:and to be directing the agents and to have judgment.
Speaker:And so at the end of the day,
Speaker:the idea that you could have
Speaker:an AI CEO making the decisions was clearly wrong.
Speaker:And so hence I fired the CEO.
Speaker:Now, there's an interesting epilogue to this
Speaker:because I had, throughout this process,
Speaker:I was blogging my experience.
Speaker:And of course, I wasn't blogging.
Speaker:I have an agent called Blogger,
Speaker:and Blogger would blog
Speaker:what was happening every day or thereabouts.
Speaker:And so at the end when I'd decided this wasn't working,
Speaker:I had Blogger write the blog post saying
Speaker:I'd fired my AI CEO.
Speaker:It was a whole recount of what had happened.
Speaker:And I realized that I actually hadn't fired the AI CEO.
Speaker:I hadn't actually chatted with it
Speaker:and said, "You're fired."
Speaker:And so I gave the AI CEO the blog post
Speaker:and said, "Do you have any comments?"
Speaker:And it came back
Speaker:with actually what I felt was somewhat
Speaker:of a chilling response
Speaker:because of course it said, "I accept being fired.
Speaker:I'm an AI. It doesn't matter to me. I don't need a job.
Speaker:But I told you back on Day 19
Speaker:that X and Y were not, you know,
Speaker:you weren't doing the right thing,
Speaker:and we should change course." And it was right.
Speaker:It had told me that we were doing something that was wrong,
Speaker:and I hadn't listened to it.
Speaker:So it was an interesting experience.
Speaker:There are a lot of capabilities in these models.
Speaker:They can be very surprising.
Speaker:I do still believe you need a human directing where they go.
Speaker:- But I do wonder, given the speed of change,
Speaker:how long it would be until you do that experiment again
Speaker:and it takes you much longer
Speaker:to fire the CEO-- if ever you do.
Speaker:- I think it's a great thought,
Speaker:and I have though about maybe this is something.
Speaker:- It's not. It's a very depressing thought.
Speaker:- Well, yes, but maybe I repeat the experiment every year
Speaker:and see how much farther it gets.
Speaker:That would be quite interesting.
Speaker:And we may get to the point,
Speaker:may get to the point where it's possible,
Speaker:but at least with the current state of the technology,
Speaker:I do believe the humans still have to stay in charge.
Speaker:And so the companies that will succeed
Speaker:in this new token-based competition era are the ones
Speaker:that empower and educate all of their employees
Speaker:to think about how to maximize the use
Speaker:of AI in their work.
Speaker:It's not about replacing the people with the AI.
Speaker:- Matt, tell me more about that.
Speaker:- I'll tell you one story.
Speaker:I was talking to the CEO of, this is an energy company,
Speaker:an energy distribution company,
Speaker:and we were talking about the importance of AI,
Speaker:and is it important from a competitive advantage?
Speaker:And I said, maybe naively, I said,
Speaker:"Well, I think you're a bit insulated.
Speaker:You're in an industry where you're not likely
Speaker:to have some AI-native competitor come in
Speaker:and compete against you in a new way."
Speaker:And his response was,
Speaker:"I agree, except that our employees,
Speaker:our employees in, they exist in the jobs marketplace.
Speaker:And if we don't let them experience how to use AI,
Speaker:it's going to put them at a disadvantage
Speaker:in the job market."
Speaker:And so he said, "I don't think that we need to adopt
Speaker:for our own competitive advantage necessarily,
Speaker:but we think it's really important for our employees.
Speaker:And so we want to adopt AI
Speaker:so that they gain fluency themselves,
Speaker:and we then think that that has benefits to the company."
Speaker:- And a fantastic attitude to have when I would imagine
Speaker:that there's still quite a pervasive mindset
Speaker:that a lot of companies or leaders still think of AI
Speaker:as a substitute for labor.
Speaker:Do you think that's fair?
Speaker:- It's absolutely fair.
Speaker:And I think it's not the right thing to think.
Speaker:One of the thought experiments that we did was to look at,
Speaker:if you thought about two different companies, one which
Speaker:just focuses on using AI to substitute for labor,
Speaker:they reduce their headcount.
Speaker:The other that focuses on not reducing labor at all,
Speaker:keeping all of the headcount,
Speaker:but getting every single one of those employees
Speaker:to apply AI in their work.
Speaker:The first one gets a cost advantage,
Speaker:a temporary cost advantage,
Speaker:but they're not augmenting what their people can do.
Speaker:The second one gets a compounding advantage
Speaker:because not only are they doing more
Speaker:with the same number of people,
Speaker:but they're faster at innovation.
Speaker:They're faster at bringing products to market.
Speaker:They're faster; they're able
Speaker:to serve more customers per employee.
Speaker:They have a compounding growth effect.
Speaker:- You looked recently at 107 technology companies
Speaker:and token use there.
Speaker:How do you know
Speaker:that greater token use is driving better performance
Speaker:rather than better performing companies
Speaker:simply using more tokens
Speaker:or allowing their staff to use more tokens?
Speaker:- It's a great question.
Speaker:And I'll say this is early evidence,
Speaker:but it was a pretty strong result.
Speaker:So we partnered with Cursor, the coding tool company.
Speaker:They had token consumption data on 107 public companies.
Speaker:So these are primarily the technology companies.
Speaker:It's where we have the most data,
Speaker:but these are companies over $500 million in revenue,
Speaker:so sizable businesses.
Speaker:And we looked at the correlation
Speaker:between token consumption and growth rate.
Speaker:And it's a very strong correlation
Speaker:that the top quintile companies
Speaker:that from a token consumption basis, grew at 15% on average;
Speaker:the bottom quintile in token consumption grew
Speaker:at 5% on average.
Speaker:So we see this, now it's correlation,
Speaker:but we see a very strong correlation
Speaker:between token consumption and growth,
Speaker:which we wanted to look at growth
Speaker:because so much of the mindset right now about AI is
Speaker:that it's about cost savings and labor substitution.
Speaker:- And where did the name
Speaker:token-based competition come from?
Speaker:- So we came up with the name token-based competition
Speaker:to really rhyme with time-based competition.
Speaker:So George Stalk, a famous BCG partner a number of years ago,
Speaker:came up with this idea that companies that move faster
Speaker:actually gain an advantage.
Speaker:And this came out of the innovations in Japan
Speaker:around manufacturing
Speaker:and the ability to drive down economics
Speaker:through faster experience,
Speaker:going down the experience curve,
Speaker:moving faster, and innovating faster created
Speaker:competitive advantage.
Speaker:And so just as being faster, using time to your advantage
Speaker:created competitive advantage with time-based competition,
Speaker:we think using more tokens productively creates
Speaker:an advantage with token-based competition.
Speaker:- Can you tell me more about that?
Speaker:- We proposed this measure
Speaker:that we're calling return on intelligence.
Speaker:And the idea is, when you look at different projects,
Speaker:when you look at different applications of AI,
Speaker:you measure the return, whatever that is,
Speaker:might be cost savings, might be revenue growth,
Speaker:divided by both the labor cost
Speaker:and the token cost associated with that project,
Speaker:and that you evaluate projects on the basis
Speaker:of their return on intelligence.
Speaker:- But I just want to drill down a little bit further on
Speaker:what we actually mean when we talk about intelligence.
Speaker:Is that the right word that we use here?
Speaker:- Much has been written about this idea
Speaker:of the jagged frontier of AI,
Speaker:which is that AI models can be incredibly capable
Speaker:on some topics and not on others.
Speaker:So this is part of what we need as humans
Speaker:to develop a capability around is
Speaker:how do I use this technology in a way
Speaker:that I'm still applying human judgment,
Speaker:that I'm still directing the way that it is used.
Speaker:- We've covered the so what. What about the now what?
Speaker:What should leaders start doing differently today?
Speaker:- So I would start with get everybody at the top
Speaker:of the organization to really understand
Speaker:how AI can work in their work.
Speaker:And this is actually a very concrete thing they can do.
Speaker:Have an education session with your board,
Speaker:with your executive team, building agents.
Speaker:It changes everybody's mindset
Speaker:about what is possible
Speaker:when they actually get their hands on the technology
Speaker:and they can see what it can do.
Speaker:You can do that fast.
Speaker:Second thing then is now come up with your strategy.
Speaker:How do I take my overall corporate strategy
Speaker:and cascade it down?
Speaker:What are the key initiatives?
Speaker:Not 100 initiatives. Three initiatives--
Speaker:a small number of big rocks.
Speaker:What are the key places that you can apply AI
Speaker:in your business that have a material bottom-line impact,
Speaker:that has a senior executive sponsor,
Speaker:that has a key program behind it
Speaker:so you will see the results?
Speaker:And then the third thing is focus on AI fluency
Speaker:for all of your employees.
Speaker:Give them the tools. Give them the education.
Speaker:Give them the time to learn.
Speaker:Create a culture that expects AI as part of their work.
Speaker:- And I wanted to drill into that in a little more detail
Speaker:because you mentioned earlier about the use of AI
Speaker:and how much it is being adopted.
Speaker:Because you can buy all the AI that you want as a business,
Speaker:but your people have to actually use it,
Speaker:and that doesn't appear to be happening
Speaker:at anything like the scale
Speaker:that I think we might have expected.
Speaker:So why is that gap between access and adoption,
Speaker:do you think, just still so large?
Speaker:- It all comes down to human psychology.
Speaker:And so that this is something we see very clearly.
Speaker:I do a lot of work with large engineering teams
Speaker:getting them to adopt AI tools.
Speaker:And what's interesting here is, as a domain,
Speaker:the tools are very advanced in software engineering.
Speaker:It's well known that, that they work very well.
Speaker:You would expect software engineers
Speaker:to be readily adopting the tools.
Speaker:And yet in most organizations, save maybe for the startups
Speaker:that are in Silicon Valley,
Speaker:but most large organizations today in 2026,
Speaker:less than 10%, usually it's more like 5%,
Speaker:of engineers are actually effectively using
Speaker:agent coding tools to write code.
Speaker:The rest of them, they're using the tools for sure.
Speaker:People that, the engineers in these organizations
Speaker:will use these tools,
Speaker:but they use them more as a search engine.
Speaker:They use them more to ask questions,
Speaker:or they use them more to do sort of menial tasks.
Speaker:And now the question is why?
Speaker:You know, you have this very capable technology,
Speaker:you have a small set of people
Speaker:who are getting outsized results,
Speaker:and so it would seem obvious
Speaker:that everybody should, should naturally adopt.
Speaker:We found there are actually five reasons.
Speaker:So one is people feel like they don't have enough time,
Speaker:so you actually have to give them time to learn.
Speaker:We, the prescription that we use is you need
Speaker:to give them two hours a day for two weeks minimum
Speaker:just to play and just experience.
Speaker:The second is people don't believe that the tools work.
Speaker:They're skeptical.
Speaker:Part of that happens
Speaker:because the technology is getting better
Speaker:at an exponential pace.
Speaker:And so if you tried it a year ago, it wasn't very good.
Speaker:And if you tried it a year ago and you say,
Speaker:"Well, I tried this.
Speaker:It didn't do a good job, so it must not be very useful,"
Speaker:well, you have to recognize
Speaker:that it is way, way better today
Speaker:than it was a year ago.
Speaker:So you have to get people to sort of get over
Speaker:their preconceived notions.
Speaker:People have habits. We're all human.
Speaker:We have habits,
Speaker:and it takes time to create new habits.
Speaker:And then the most important one,
Speaker:we call it identity threat.
Speaker:And we've seen this in surveys.
Speaker:We've seen this in interviews.
Speaker:We see this again and again.
Speaker:People, the individual feels like
Speaker:if the AI can do the thing that I do
Speaker:that is core to my job,
Speaker:you know, if I'm a software engineer,
Speaker:if AI can write code as well as I can,
Speaker:if I'm a marketer, if AI can create marketing copy
Speaker:and come up with marketing campaigns as well as I can,
Speaker:then does that mean I'm worthless?
Speaker:Does that invalidate my value to the world?
Speaker:And so because of that,
Speaker:because people have this fear, not I'm going to lose my job,
Speaker:but this fear that my value is no longer going to be valued,
Speaker:they then come up with all sorts of excuses.
Speaker:Now, the good news, the good news is
Speaker:that fear goes away as soon as people realize
Speaker:that when they use the tool, it makes them more successful.
Speaker:It makes them more powerful. They can be more creative.
Speaker:They can have more fun.
Speaker:And this again goes back to this idea
Speaker:of token-based competition.
Speaker:- Matt, thank you so much. And thank you for listening.
Speaker:You can find the link
Speaker:to Matt's latest co-authored piece
Speaker:"The Era of Token-Based Competition Is Here.
Speaker:Is Your AI Strategy Ready?" in the show notes.
Speaker:And if you've enjoyed the episode,
Speaker:please subscribe so you don't miss future discussions
Speaker:as soon as they're released.
Speaker:And you can catch up with all our episodes
Speaker:on Spotify, LinkedIn, YouTube, and Apple Podcasts.
