In this episode, the compounding productivity gains from AI workflows take center stage, framed against a debate over whether human talent now breaks even against raw token costs. The argument lands on a double-factor effect (models improving while operator skill improves week over week) and pulls in Mark Benioff’s view that a middleware layer will soon sit between LLMs and end users to stop tokens being wasted rebuilding work from scratch. Serious Land Capital ties it back to its own roadmap, moving deal-comp analysis from a standalone web app toward an internal cowork-style system.
Key Takeaways:
- Productivity compounds on two axes Output gains are accelerating because the frontier models improve weekly AND operator skill at using them improves weekly, a double-factored effect that is unmistakable even when it does not yet hit the bottom line.
- Token-cost comparisons are context-dependent Reported break-even metrics ignore whether teams are trained, on subsidized plans versus raw API, and using frontier US models versus open-source Chinese LLMs that run 10 to 30x cheaper but sacrifice top-tier intelligence.
- A middleware layer will ration tokens Benioff’s insight: companies will sit between the LLMs and end users as a sieve, pulling proven frameworks from shared databases so a keynote or build is not regenerated from scratch by millions of users at once, the same way Squarespace and WordPress templates replaced building from zero.
- Enterprise SaaS endures, vibe-coded apps hit a ceiling Dominant players reinvest heavily into AI to improve already-winning products, while smaller vibe-coded tools break under load (Callan’s Accelerate to Automate launch broke Zapier and Zoom integrations and forced an emergency move to enterprise-grade rails).
- Serve the customer, full stop Benioff’s directive through any SaaS-pocalypse is to keep delivering value, and Levesque’s product-market-fit Venn diagram reinforces it: solve your own problem and a large audience almost certainly shares it, which is why the SLC plugin is being rebuilt to pull comps faster and more automatically.
If you want the full reasoning on where the token economy is heading and what it means for builders, listen to the complete episode.
(Podcast transcript below)
Welcome to Get Serious. We’re at Serious Land Capital. We have funded over six and a quarter million dollars worth of land deals with industry leading 41% operating margins. So you know, it’s a personal update here. Some of you who might have read my newsletter or heard you know or seen over social media. my wife and I welcomed a newborn, our little baby boy Zende over the past week.
if you’re listening to this in in early June. So we are absolutely under fire with nonstop action back at home along with our our three and a half year old daughter yeah trying to take charge of a lot of childcare duties to the best of her ability as well too. So you know the not not the best sleep at the moment a as any parent can relate and you know logistically challenged but you know
back in the driver’s seat and slowly but steadily getting things back into motion here. nevertheless, you know, professional obligations never stop right and, you know, thought processes and what’s going on in the world, like they’re just again never been such a fast moving environment. So I wanted to share a few other updates, both in relation to
what we’re doing, and just some you know general observations and of course, you know, kind of touch on the AI side, where again, yeah, if you’re paying attention to Claude, Opus 4.8 just dropped, and even more tools on you know how to increase the utilization of of tokens and you know i improve throughput even more than what was possible again, even compared to a week ago here.
and I’ve seen some narratives where you know, initially it was like, my gosh, look, you know, how much more work can be done utilizing AI workflows compared to human talent. and I I’m still firmly in that camp for most tasks that are well suited for AI workflows. yeah, again, I would think of a lot of, you know.
regulatory items or legal work or yeah drafting almost anything written or you know reviewing deeper details about I’m I mean we’re we’re we’re kind of going over everything. Like it it I I’ve shown time and time again you can utilize AI for like almost any problem and you can always break the problem down further to at least get get some progress.
But you know, the takeaway here is that in our experience, you know, my productivity and my own team’s productivity has just accelerated and continues to accelerate on a week by week basis through both the AI models continuing to get better as well as our skill set at using AI continuing to improve week over week. So you kind of get a a double-factored paradigm there leading to these gains.
And like it it it’s unmistakable how much our our productivity has has increased. You know, again, not always going to show up in an immediate bottom line situation, because again, you have to account for inherent market conditions and so forth and you know, risk off tendency and you know, real estate on average is you know slower than than it had been. But in several other metrics, just
In terms of like speed to produce content and distribute and you know increase site visits and deal flow and so forth, like we are doing much, much better from from that perspective. and sometimes like you just have to look at it subjectively, like, yeah, am I able to have greater output than I was, you know, a year ago? Like the the answer is unmistakably yes, unmistakably. so just bearing that.
in mind here. But like I’ve seen some of these metrics where it’s like, no, you know, some of the, you know, human talent is basically break even compared to using token costs or you know, to tokens for AI models. some it might even be a little bit worse here. And like we just have to bear in mind like you know what are these companies that are reporting the these usages? You know, how trained are people
in utilizing the AI, like do they all have you know more of the subsidized subscription plans through cloud, or they having to buy you know raw API tokens? Like the latter is certainly going to be more expensive here. So it’s it’s contextual, just like any question you ask on what the actual reality is. there’s no doubt that the US models, mean, think the open AIs and Anthropic and Gemini and so forth, like those cost more.
compared to other models. And a lot of people will think of the open source Chinese LLMs that are, you know, 10 to 30x less expensive, but they’re not the frontier models either. So, you know, you can see some of the metrics where companies that are just have extraordinary engineering challenges are utilizing the open source Chinese models instead due to cost reasons. but it’s at the sacrifice of not having.
the highest level of intelligence possible. So, you know, it depends on what your trade-off is is going to be in that regard. So to me, I I think it’s kind of a two two factor piece here. One is that I think a lot of companies are just wasteful in their approach to figuring out how how to best use tokens. and
Yeah, to a certain degree too, like, you know, even if it’s a shorter term cost, again, it’s a little bit of a tangent here, but if you can get to an answer faster or serve your market even faster, like is that going to be worth the trade-off? sure, of course, you gotta manage your cash flow and not not let yourself sink under. But if you can increase your medium to longer term moat and, you know, continue to pace ahead of your competitors through, you know, just
extraordinary usage and experimentation with ai tokens, like that trade-off might be better in in the long run. so that’s one piece. And then the other piece, which I I I thought was just, you know, brilliant, was hearing Mark Benioff speak about this within the past few weeks. and this is just a reminder too is like, you know, th there’s there’s always levels to the game. And you know, learning from some of the
sharpest minds in in tech who’ve been doing this for decades. And you know, bending off obviously multi-billionaire Salesforce is gosh, they’re like a hundred and fifty billion dollar market cap company. la last I looked here, you know, bringing in close to like 50 mil in in annual revenue and sixteen, twenty mil of of cash flow annually, like monster numbers, right? and you know, they’re burning
$300 million worth of Claude tokens on an annual basis here. Again, like way beyond most companies can even consider. And in relation to like their total company size, you might even say like that’s not as much as they could be doing. but still tremendous, tremendous amounts here. and so like company large companies like that are clearly putting these models through the paces. but what Mark was mentioning too is like, yeah, the you know, cost per token is definitely.
significant. but he doesn’t see that being a longer term issue, and that right now it’s just a bit messy on how the AI models are distributing tokens for the work that’s being called upon them. And most illustratively about this is like anytime I don’t know, you dive into cloud cowork cloud code or do any type of vibe coding project or you know, not even necessarily a
I mean, all AI kind of uses is, you know, some inherent coding in in the background. Like it has to you know ha have some type of so software throughput to arrive at at an answer, regardless of of the query that you’re you’re presenting it. but you know, pretty much always you’re having to start or you know, the LLM is having to start from scratch.
Yeah, and maybe not always from scratch if it’s utilizing infrastructure that you’ve already built within your business and it’s able to call upon work that had previously been done. But each you know, business or user of AI and even within business, like, you know, me using cowork versus my team member using cowork, we’re gonna be in siloed areas. So like it’s not necessarily shared in in that regard here. So like a lot of work.
might be redundant or you know repeated unnecessarily and just building things from scratch. Like there could be, you know, a hundred million people at once asking, hey, I want to build like this, you know, keynote presentation utilizing this type of design structure and so forth too. And like, you know, Claude or OpenAI, like throw in an E L L like they’re going to start from scratch to build that. but the thing is, is once certain
You know, call call it frameworks, just like a lot of the existing software engineering languages, as people would make progress in building certain pieces of infrastructure, like those building blocks would then be effectively publicly available or at least you know privately available through whatever company has access to the the source code. And then, you know, whatever engineer is working on a particular project, if there’s something that
can be utilized across multiple different projects or multiple different product lines, then you don’t need to rebuild that from scratch. So again, like th think from, you we’re we’re in real estate, right? So, you know, building a house here, like if if somebody builds a front door that, you know, can can fit well with you know a whole bunch of other houses that might be different square foot square footages and and so forth.
but the door frame is the exact same, then yeah, I I I just need one person to create that door. And maybe you can throw some different different aesthetics onto it, but like the raw material, hey, I need this type of wood cut cut down into the this specific size. that that shouldn’t have to be done all the time here. So what Mark Benioff was was mentioning is
Yeah. He he sees that there’s going to be some companies that sit in between the LLMs and end users that will basically filter you know how many tokens need to be distributed to a particular project that that serves as kind of like a a sieve or a sieve you know tomato tomato type situation that it’s like okay, I don’t need just a gargantuan amount of tokens thrown at this thing. Like I need to pick and choose.
precisely what needs to be utilized for this particular project. And I can pull from maybe some public database, even private database, what have you, of frameworks that have already proven to work for particular projects. So like, yeah, all of a sudden, hey, there, yeah, PowerPoint keynote, presentation, what have you, the this framework’s already been done. I don’t need to burn, you know, 100,000 tokens, rebuilding that from scratch, you 100 million people all around the world doing the same
Same thing. again, these numbers are illustrative, right? and so it just becomes a lot more efficient in terms of how we are utilizing the compute that is available because that’s like always gonna be an in-demand, right? And you know, difficult energy situation around the world right now, and everybody just like demand, demand, demand for these these tokens.
to to create work on the back end. so to me, like I I think this is just inevitable. Like we’ve what we’ve already seen, like it’s not like this hasn’t happened. Like I mentioned with previous software languages, people just built frameworks all the time. And so you could start. I mean to think about if you use Squarespace or WordPress and all that, like there was like in infinite templates to choose from. Like those were frameworks that somebody else had already created. So you didn’t have to build it from from scratch. same thing
I I think we should be able to expect within AI creation of certain projects or vibe coding and and so forth here that will just make everything much cheaper and faster to to do going forward here. Now, when that might happen, I don’t know. but certainly some of these top tech leaders are thinking about it and it’s like a an obvious problem to to help solve. so I I just thought that was a really, really brilliant
insight and also like just and lending some some further thought okay like what needs to what needs to evolve here and again just showing okay what what are the true tech titans you know how are they utilizing the these technologies and you know potential directions it can go into so yeah again that that was just more you know even
a humbling moment for me too, where it’s like, yeah, that some of these people are just so brilliant and and the levels of thinking that that they’re engaging in is just h helping reframe my own conceptualization of these particular issues. Another note too is you know we’ve heard a lot of like the death of SaaS products and like is software even investable anymore still TBD long term, right?
but you know, d d during that particular conversation I was seeing with Mark Benioff and just seeing some of this in in real time here is that you know pr th those top players, like the real enterprise softwares, again, you know, think like the Microsoft Office tools or yeah, Salesforce that’s you know ha been able to serve millions and millions of clients over a long period of time. they will likely have a longer or a better chance to succeed. First because like they’re again throwing
Hundreds millions of dollars, if not more, toward utilizing these AI tools to improve their own products. So like again, they they have the benefit of being the grill in the room to improve an already dominant product. but another piece is that you know, a lot of these smaller vibe coding products or, you know, anything that might be built on like a replit or a lovable or what have you, like the the underlying AI models are not necessarily going to help with.
bandwidth. so you know that this is something we we saw internally at Callins Uncommon Business. Like they just you know gained way more customers than they were expecting this last launch for Accelerate to Automate. and I’m help helping advise their company now too. So just see seeing a lot more behind the scenes and like a lot of their systems, you know, Zapier, how you know add events, like being able to add calendar events to clients calendars and and so forth, even Zoom.
Like it all broke on them. and so they had to just rapidly you know, either switch over to more enterprise grade tools or just upgrade to enterprise plans in general. So like that that’s just another demonstration of, you know, okay, yeah, you you might be able to b spin up some, you know, nifty software really quick, but like if you’re not able to distribute it effectively and have the rails in place to
allow for a significant volume of users, like things will fall apart eventually as well too. And like I just don’t see yeah, if you don’t have the hosting power, like the ability to handle that bandwidth, which is gonna require its own capital and resources anyway. like a lot of these vibe coding software projects are you know more relegated to smaller, you know, they’re they’re kind of capped from a user perspective.
or they’re just gonna have to be, you know, invested in too much further to be able to serve larger clients. So like to me, that that was just another you know key thing to to point out here is that yeah, enterprise software, like the good stuff, is probably here to stay, especially the ones that continue to you know take advantage of all the frontier models for for for AI to to improve their their own product. And like that was.
You know, the to to end on this, like that was Mark’s core takeaway. Here is like, you know, regardless of the SAS pocalypse or then he’s like, Yeah, this is not the first time, you know, SAS products have been hit. but like during all of this, you know, regardless of what the market cap shows, the stock price, like, yeah, you’re you’re not gonna be able to control the market. but all he focuses on and all he tells his team to focus on is like just keep delivering value to your customers. Like that, and we we hear that all the time, right? It almost sounds trite, but
again, that this is like a multi billion dollar market cap company that’s you know proven itself over many years here and has real staying power just because they they keep focusing on the customer, keep focusing on the customer. And like I see that in Callan’s business as well, too. And you know, we we try to do that at Sirius Land Capital, and like all the best companies out there, right? Like they they serve the customer better than than their competitors.
And so one one of my other mentors, Ryan Levesque, I’ve mentioned him as well, too. He was huge in the internet marketing space, now does a lot more, you know, smaller consulting. but you know, has some brilliant opinions on AI and so forth, new new book coming out as well here, too. But I loved one of his recent illustrations where it was just a Venn diagram of, you know, when you’re trying to find product market fit. It’s like, you know, whatever core problem you have.
almost assuredly there’s significant overlap with with what your customer base or potential customer base is you know, in need of a solution of as well too. So it’s like, you know, solve your own problems and almost assuredly a bunch of other people are going to have the same or a very similar problem. So like that’s what I’m trying to focus on, and like the you know, SLC chat cowork plug-in, you know, just
Based on even internal, external feedback, like, yeah, it just has to be faster. like I I need to make it more automated, be able to pull up comps through some of these pathways, maybe even through a cloud code setup instead of cloud cowork. Doesn’t really matter. like the feedback is just so much faster to to get nowadays. and it’s like, okay, it within Sirius Land Capital, like yeah, I want to be able to throw more deals in into a system like this, just you know, have it go to work.
figure out, okay, get me get me the right comps to start off with. And then we can we we can make some further assessments from there based on that initial analysis. And so like we’re gonna be building that in internally here too. Like the the web app was a good first start as kind of a test case and like learning how yeah how did how do these API tokens work? Like what what’s the best way to serve the customer. and now yeah tr trending in that that cowork direction because it’s like okay, yeah, this would
serve my own needs needs better and it would serve the the needs of of our audience better. And so you just you have to find a way to serve value to the customer to you know succeed in the market and like that’s the ultimate takeaway regardless of what technologies are on hand or how much it costs to utilize any of these things. So hopefully some of these takeaways I know a little bit more technical here and there but like they are massive topics for
you know, the state of today’s industry and kind of holding up our stock market. So I I think it’s important to note you know, some of the larger narratives and what what the truth might actually be out there and what to look forward to. So with all of that, subscribe and share, everybody. looking forward to talking to you all next time. And if you have a deal to submit, seriousland dot capital, send it over. We will take a look straight away. Take care.
