Tenex's token spend is roughly one to two times employee spend

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 14:43

   this year, is percent of let's say token spend to employee spend. I don't know if you have that number. So roughly, most companies are at maybe let's call it $10 to $200 in spend per employee. The more AI native ones would convert that to a percentage, so roughly 10%. The average AI startup UpEmployee is about 1x in terms of employee spend and token spend. And then it goes up to maybe 5 to 10x in terms of token spend.

2. 15:15

   Where are you guys in that range? I actually don't have a number off the top of my head. I'd say we're somewhere between, if we're talking like raw tokens, somewhere between 1 to 1 and maybe like 2 to 1. Really? That's pretty high. And what we think about a lot is like when to reach for the biggest model. I think every company is starting to think about that. But I will say recently our costs have been driven just way down from Opus 5.5.

3. 15:45

   We kind of had an era of like everyone was using Astra and Fable and they just gave the best results. But ultimately those models are expensive. And like a lot of like, you know, since the last time we talked, a lot of people had this sort of token maxing error, token leaderboards, everything. And then they were like, oh, no, no, no, no. Now we have too much cost. And like, well, it's the most predictable thing in the world.

4. 16:03

   Did you ever fear about it? or like you were just getting good ROI regardless? Yeah, I mean, there's a few ways that I think about it. It's like we were for sure worried about it just as like our headcount was growing. And I think also like even though our team, I would say like I don't feel like anyone on our team ever really ascribed to like the token maxing zeitgeist because... I had a few moments.

5. 16:35

   Yeah, me too. But it's like, well, if the boss is going to pay for it, I'm just going to go nuts, right? Because then you do the weird stuff so that you can scale back. But you have to do the weird stuff first to discover the lessons that you can take with you. Yeah, totally. I mean, this may sound crazy, and I'd be curious of your thoughts, but I was talking to someone earlier about the debate that continues around, basically, I promise this is related, I was talking to someone earlier around the data companies, right like um micro one record handshake and just like what is like what does the long term for these businesses look like right because like not 90 plus percentage of their their revenue is from the labs and a number of these founders would argue that they think it's going to flip-flop

6. 17:21

   where in the next few years it's going to be enterprise driven and then like the thing that comes to mind like the next thing i assume you have to believe if enterprises are going to be the core buyer of that data is that enterprises are going to be doing a lot of fine-tuning on their own models. And then the question becomes, where is that going to make sense? It's either you think there's going to be better quality, which again, like the bar is just going to keep moving on the frontier. So that's like a hard bet to make. Then it has to be price. The price one feels hard to me because it's like, we're just seeing token costs continue to go down on these models. And so I guess part of my, the reason we're not concerned internally is because it It almost feels like we've kind of hit this like Pareto point of like costs are coming

7. 18:06

   down at a per task basis. Then it is way harder today for someone to token max and like really F the company in terms of cost than it was six to eight months ago. Like almost think six to eight months ago was the point at which like you'd be really concerned. Yeah. Do you agree with that? Yeah. I think there's a couple ways like the future might play out there. Like, if there's a world where there's a Astra 6.5, a Fable 5.5, and they're super expensive, but they get the best results, I could see us trending back that direction.

8. 18:38

   But generally, I agree that, you know, you look at something like Muse Wonder that they launched yesterday, today, and it's on the frontier. It's dirt cheap. The competition is heating up, both in open source and closed source. And I think that just will continue to drive costs down. And when you really look at it, a lot of tasks that are economically valuable can be done with Opus 5.5. So, you know, it's that cliff of like, would a 10 times smarter model really justify the cost of like moving some cells in a spreadsheet?

9. 19:14

   Probably not. Right. So coming back to your customers and what you're seeing from the market, what is the most common ask that you guys get? And then what is something that people come to you with and you're like, no, we don't actually want that job yet?