WEBVTT
00:00:01.600 --> 00:00:11.359
For the past three and a half years, AI has almost felt too cheap to be real, especially for power users during early 2026.
00:00:11.679 --> 00:00:27.760
I mean, you pay a monthly fee, you open Chat GPT or Claude or Codex, Gemini Copilot, whatever, and then you have agents run for hours completing your actual work with your contacts without paying an additional penny on top of that subscription.
00:00:28.079 --> 00:00:33.840
But that version of AI is ending and it's gonna hit even harder later today.
00:00:34.000 --> 00:00:34.479
Why?
00:00:34.719 --> 00:00:43.759
Well, that's because Anthropic's Fable 5 only has hours left until it's no longer included in paid subscription plans.
00:00:44.000 --> 00:00:46.960
And that's not just an anthropic story.
00:00:47.119 --> 00:00:52.079
And it's the clearest signal yet that Frontier AI is becoming a metered utility.
00:00:52.240 --> 00:00:56.240
I mean, Microsoft GitHub moved co-pilot into AI credits.
00:00:56.399 --> 00:01:14.719
Google changed Gemini limits around compute complexity features and chat length, and Microsoft started pricing co-pilot cowork through credits, not included subscriptions, and even NeoClaude XAI's Grok has been moving users toward share weekly pools and extra usage credits.
00:01:14.959 --> 00:01:25.920
So with the explosion of agentic models that can now work for hours with expert level artifacts, the question is no longer how do we get everyone in our company using AI?
00:01:26.079 --> 00:01:32.159
The board level question is how do we use more AI without bankrupting our budget in the process?
00:01:32.319 --> 00:01:35.200
And that's exactly what we're going to be tackling today.
00:01:35.439 --> 00:01:37.359
So here is the big picture.
00:01:37.599 --> 00:01:41.359
Frontier AI has definitely now become a metered utility.
00:01:41.599 --> 00:01:54.640
And the wild west days of using whatever model you wanted around the clock and just paying a$20 or$200 a month subscription fee and nothing more, those days are pretty much gone.
00:01:54.879 --> 00:02:01.519
Um like I said, for three years we've been doing this seemingly unlimited use of AI.
00:02:01.599 --> 00:02:13.120
And if you go back to literally back in 2023, even when the models were pre-agentic, I even said eventually we're gonna see 200,$2,000 subscription plans.
00:02:13.199 --> 00:02:15.599
And I think even at that time, it's still gonna be a steal.
00:02:15.840 --> 00:02:28.479
And now when I throw some numbers out, you're gonna see that even these$200 a month plans, or if there isn't even a plan on top of that someday, will still be a steal considering what you're gonna start paying tomorrow.
00:02:28.639 --> 00:02:44.159
That's because even though the price per token has fallen by literally like 98% over the past three years, depending on what tier you're looking at, the agentic capabilities and the amount of tokens that these models are using has legit skyrocketed.
00:02:44.319 --> 00:02:53.360
And over the past two or three months, AI has quickly gone from let's use as much as we can in token max to putting strict spending limits.
00:02:53.520 --> 00:03:03.520
And sometimes, you know, developers at huge companies aren't even to use AI as much as everyday non-technical people like you and me.
00:03:03.599 --> 00:03:14.000
And I think the fable issue today in Thropix Fable V leaving subscription uh in a couple of hours only heightens that chatbot bill problem.
00:03:14.159 --> 00:03:16.319
So on today's show, here's what you're gonna learn.
00:03:16.400 --> 00:03:25.360
You're gonna learn why Fable Five subscription exit is the cleanest and loudest warning shot for subscription AI and your team's AI strategy.
00:03:25.439 --> 00:03:31.439
You're gonna know why GitHub, Copilot, and Grok swapped unlimited access for pay-per-use meters.
00:03:31.520 --> 00:03:37.680
You're gonna understand why companies are capping spend, but still trying to use more tokens and how they're doing it.
00:03:37.840 --> 00:03:44.000
And I'm gonna leave you at the very end with our seven-step playbook to control runaway AI spend.
00:03:44.159 --> 00:03:45.199
Let's get into it.
00:03:45.439 --> 00:03:48.960
Welcome to Everyday AI and our Start Here series.
00:03:49.120 --> 00:03:55.280
This is the essential podcast series to both learn the AI basics and to double down on your AI knowledge.
00:03:55.520 --> 00:04:01.840
So if this is helpful and if that's exactly what you're trying to do, make sure you go to starthierseries.com.
00:04:01.919 --> 00:04:06.000
That is going to give you exclusive access to our inner circle community.
00:04:06.159 --> 00:04:20.079
You're gonna be thrust straight into our Start Here Series space where you can go listen uh to every single Start Here series, the simplest way to do it on the playlist that we have there, as well as you can go read about it, connect with other people who are trying to do the same thing.
00:04:20.319 --> 00:04:33.040
And if you missed our last Start Here series episode, we talked about the desktop agent lingo simplified, going over goals, loops, plans, sub agents, and how it works in codex and clawed code.
00:04:33.120 --> 00:04:49.120
And well, if you're doing all of those things, chances are yeah, your bills might start getting more and more expensive, which brings us to today's episode AI cost control 101 and why your chat bot bill is now becoming a board level problem.
00:04:49.759 --> 00:04:56.639
So over the past couple of months, we've seen this, and it's not just Anthropic's Fable Five.
00:04:56.800 --> 00:04:59.920
I mean, let's go back to May, right after Google's I.O.
00:05:00.079 --> 00:05:11.040
conference, their big yearly conference, for the first time, Google started putting actually pretty strict restrictions uh on their Google Gemini.
00:05:11.279 --> 00:05:18.160
So it's actually the first time I'd ever uh run into Google Gemini limits on their$20 plan.
00:05:18.319 --> 00:05:20.639
Up until then, it had been pretty generous.
00:05:20.800 --> 00:05:36.319
And, you know, after they made this shift of actually starting to count longer, uh, you know, taking things like complexity, the model you use, and chat length, all of a sudden that$20 plan that used to feel pretty great started feeling not super useful.
00:05:36.480 --> 00:05:45.279
Uh, then in June, you had Microsoft's GitHub Copilot replaced their premium request with token-based AI credits.
00:05:45.439 --> 00:05:49.759
So the unlimited plan was gone for Microsoft's GitHub Copilot.
00:05:49.920 --> 00:05:52.079
The same thing with Copilot Cowork.
00:05:52.240 --> 00:06:02.639
Microsoft is starting to go with task-level credit pricing and no longer including essentially unlimited co-work usage after they moved it to general availability.
00:06:02.720 --> 00:06:12.639
Uh, then a couple days ago, we also saw Grok uh from XAI Space XAI, I think is technically their new name, as of a couple hours ago.
00:06:12.720 --> 00:06:22.240
Uh, right, they started rolling out uh the credit base system as well, which is pretty telling considering, well, they're kind of like a neo-cloud leader now, right?
00:06:22.319 --> 00:06:28.800
They are uh one of the companies that has technically extra compute and they're even charging for it.
00:06:28.959 --> 00:06:39.279
And like I've already said, today is maybe the day that all of this gets thrust more squarely into focus because of Anthropic's Fable 5.
00:06:39.439 --> 00:06:41.120
And why is that important?
00:06:41.360 --> 00:07:01.040
Well, right now, at least, I mean, we'll see what happens over the next couple of hours because we've heard uh rumors that we might get a GPT-5-6 from OpenAI today, uh, sometime between today and Thursday, but uh, at least for the next couple of hours, uh, Fable Five is the most powerful AI model in the world.
00:07:01.199 --> 00:07:05.920
And at least for a couple more hours, it is still available in a subscription.
00:07:06.000 --> 00:07:10.720
But after today, you are going to be paying the full API price for that.
00:07:11.120 --> 00:07:20.800
So we've gone over the uh Fable Five drama kind of enough, but you know, they put it out there, they had to take it away because of some problems with the US government.
00:07:20.959 --> 00:07:27.360
Uh, and they brought it back a couple of days ago, but they said only included in plans through July 7th.
00:07:27.439 --> 00:07:32.000
And they said, hey, maybe one day when we have enough compute, maybe we'll bring it back, but no guarantee.
00:07:32.160 --> 00:07:33.600
And why does this matter?
00:07:33.839 --> 00:07:43.199
Because, like I said, for the past three and a half years, we've been able to, you know, whether you have a$20 plan,$100 pro max plan,$200 pro max, right?
00:07:43.360 --> 00:07:48.000
But for a very low price, and I know that's all relative, right?
00:07:48.160 --> 00:07:55.040
For many people out there, I understand a$200 a month plan, especially if you're paying it for it yourself, doesn't seem like the cheapest thing.
00:07:55.199 --> 00:07:57.519
But let me just show you why it actually is.
00:07:57.839 --> 00:08:02.560
So now, starting tomorrow, uh technically 11:59 p.m.
00:08:02.879 --> 00:08:12.079
Pacific time, you're gonna be paying uh$10 per million input tokens and$50 per million output tokens for uh Fable 5.
00:08:12.560 --> 00:08:14.800
Let me just put this into comparison now.
00:08:14.959 --> 00:08:17.360
So you're probably saying, all right, what's the big deal?
00:08:17.519 --> 00:08:31.439
All right, like Jordan, if you're paying, you know,$200 uh, you know, a month for, you know, which I am, I'm paying$200 a month uh for the Claude 20X Max plan, I'm paying$200 a month for the ChatGPT Pro plan.
00:08:31.680 --> 00:08:37.360
Uh, you know, I paid the$200 a month previously for the Google uh Ultra plan, so I paid for all of these.
00:08:37.519 --> 00:08:39.360
So you're saying, okay, what's the big deal?
00:08:39.519 --> 00:08:44.480
You know, if it just moves to API only, can't you just use that money for the API credits?
00:08:44.799 --> 00:08:45.679
Yeah, no.
00:08:46.000 --> 00:08:46.320
Okay.
00:08:46.799 --> 00:08:53.519
So granted, OpenAI has still ridiculously generous limits.
00:08:53.679 --> 00:08:54.000
All right.
00:08:54.159 --> 00:08:55.519
So let me put that out there.
00:08:55.679 --> 00:09:12.000
You're normally not getting that level of spend on these other plans, but I've been averaging uh over the past couple of weeks about two-ish billion tokens a week on the$200 a month uh OpenAI codex plan.
00:09:12.320 --> 00:09:15.279
Okay, uh, which is included with ChatGPT, right?
00:09:15.360 --> 00:09:16.159
So it's the same plan.
00:09:16.480 --> 00:09:21.200
Uh so that at a at a very generous blended rate, right?
00:09:21.279 --> 00:09:32.639
Of uh a blended rate of input and output tokens, that is a$200,000 a month anthropic bill.
00:09:32.879 --> 00:09:48.799
If I were to take everything I'm doing inside of Codex and I'm only paying$200 a month, and if I were to say, in theory, move all of that into um Fable, that is$200,000 a month.
00:09:49.840 --> 00:10:01.679
Okay, now you see why this has been a such a big discussion in the AI world about no longer having this democratized AI uh access to AI.
00:10:01.840 --> 00:10:06.399
We'll see uh, you know, what it means with uh OpenAI's models moving forward.
00:10:06.480 --> 00:10:18.799
Uh, they did say that their, you know, new GPT-5.6, uh kind of the different tiers, uh, Seoul, Terra, Luna, and you know, Soul Ultra would be included in subscription plans.
00:10:18.960 --> 00:10:35.360
So maybe OpenAI may be one of the only few players left that has yet to um, you know, kind of knock, knock out that subsidy, but that is still the reality uh because you have to assume that one day that's gonna be the status quo.
00:10:35.440 --> 00:10:41.360
And maybe open AI might be able to do it for another three months, three quarters, three years.
00:10:41.519 --> 00:10:42.399
We're not sure.
00:10:42.639 --> 00:10:52.159
But eventually, AI is going to get more and more expensive because the models themselves, in theory, are going to be eating up more and more tokens.
00:10:52.399 --> 00:10:57.200
And this is exactly why we're seeing that token maxing to token efficiency shift.
00:10:57.279 --> 00:11:02.080
And we did cover that on the start here series, volume 27 or episode 789.
00:11:02.159 --> 00:11:03.200
So go check that out.
00:11:03.279 --> 00:11:05.919
But we've seen it in some of the biggest companies in the world.
00:11:06.159 --> 00:11:16.879
So, as an example, uh, we saw the very uh you know popular story, Uber reportedly burned through its 2026 AI coding budget in just four months.
00:11:17.120 --> 00:11:22.799
Tesla, we recently saw capped employee AI tool spend to just$200 a week.
00:11:22.960 --> 00:11:30.559
And then UBS says that 60% of interviewed enterprises are throttling AI spent already, right?
00:11:30.720 --> 00:11:35.519
So I do assume that number will probably go up if UBS does the same survey quarterly.
00:11:35.600 --> 00:11:36.480
I don't know if they do.
00:11:36.720 --> 00:11:51.679
I would assume that's gonna go up to about 70 to 80 percent uh by the end of the third quarter, because I don't think the uh kind of the token efficiency uh wall has hit the majority of enterprises just yet.
00:11:52.000 --> 00:11:56.000
So, and this is actually everywhere, right?
00:11:56.080 --> 00:12:06.720
I even had a conversation with uh a friend who is in software development at one of the Mag 7 companies, you know, and he said his spend is$100 a month, right?
00:12:06.799 --> 00:12:09.279
So that$200 a week at Tesla.
00:12:09.440 --> 00:12:12.320
Although when you're looking at API credits, you're like, that's like nothing.
00:12:12.480 --> 00:12:15.919
Yeah, imagine having like$100 a month.
00:12:16.000 --> 00:12:17.120
So that is the reality.
00:12:17.279 --> 00:12:27.440
So I also want to tell this to people out there who maybe don't talk to a lot of people at Fortune 100 type companies and you know, big mag 7 type companies.
00:12:27.519 --> 00:12:32.480
We are seeing actual real spend limits on the amount of AI that these employees can use.
00:12:32.639 --> 00:12:40.799
And that actually, I think, for SMBs and medium-sized enterprises might create a strategic advantage, right?
00:12:40.879 --> 00:12:50.799
If the biggest players are maybe uh, you know, having to reel in their spend, and maybe part of that was their own fault for pushing these token leaderboards and all these other things.
00:12:50.960 --> 00:12:54.960
I think there is an advantage to be had for everyone else.
00:12:55.279 --> 00:12:59.759
All right, let's take a quick break for a word from our partners.
00:13:00.240 --> 00:13:05.279
Here's what most AI tools still can't do: work outside their own little box.
00:13:05.440 --> 00:13:07.759
The all new Slack bot just changed that.
00:13:07.840 --> 00:13:14.720
It's your AI teammate inside Slack, and now it can read, write, and act across the other apps your team already uses.
00:13:14.879 --> 00:13:20.240
No more tool switching, no more re-explaining yourself every time you open a new tab.
00:13:20.399 --> 00:13:26.879
One ops team at engine says the summary feature alone saves them 15 to 20 minutes of use.
00:13:27.039 --> 00:13:30.320
See what Slackbot can do at Slack.com.
00:13:32.720 --> 00:13:34.799
Okay, so why now?
00:13:35.279 --> 00:13:35.600
Right?
00:13:35.759 --> 00:13:42.080
Why are we seeing this uh, you know, your chatbot bill becoming a boardroom problem now?
00:13:42.240 --> 00:13:44.000
It sure wasn't at this time last year.
00:13:44.080 --> 00:13:46.080
It wasn't really at the end of 2025.
00:13:46.240 --> 00:13:55.440
And I think to make a very short story of the last six to eight months uh of AI developments, if you listen to the show, it's no surprise.
00:13:55.600 --> 00:13:57.519
Maybe if you're newer here, let me fill you in.
00:13:57.679 --> 00:14:06.559
And Tropic set the pace at the end of 2025 uh with Claude Code, bringing that to the desktop uh as well as Claude Co-work in early 2026.
00:14:06.639 --> 00:14:12.240
And then I think the rest of the way has been blazed uh by Codecs, Cursor, and some others.
00:14:12.480 --> 00:14:18.159
But we are now having extremely powerful AI models, even for non-technical work, right?
00:14:18.240 --> 00:14:24.480
When I talked about spending, you know, two-ish billion tokens a week, I'm not doing a ton of traditional coding.
00:14:24.559 --> 00:14:33.440
Yes, I'm building a bunch of programs that I use all the time, but a lot of my token spend is just doing right scheduled work that I wouldn't normally be doing, right?
00:14:33.519 --> 00:14:36.720
Doing a bunch of research, creating a bunch of just documents.
00:14:36.879 --> 00:14:46.080
Um, and the reason why we are seeing this explosion is because these capabilities are becoming more powerful and they're becoming easier for non-technical people to use.
00:14:46.240 --> 00:14:55.120
So you have more and more people starting to automate their everyday work with these desktop agents that can work 24-7 and can work for hours, right?
00:14:55.279 --> 00:15:04.159
I remember even in the earlier days in 2026 with Claude Code, I would have to really push the models uh to work for maybe like an hour or so.
00:15:04.320 --> 00:15:20.000
But now with the combination of subagents and plan mode and goal and all those things that we talked about in the last start here series, every single day I'm usually having multiple threads running for eight to 16 hours and creating amazing outcomes.
00:15:20.320 --> 00:15:24.960
So it's all about the amount of tokens being spent.
00:15:25.200 --> 00:15:37.440
So a Stanford Digital Economy Lab study this year showed how much more impactful and how much more token hungry today's models are than the old chatbot only non-reasoning models.
00:15:37.600 --> 00:15:48.320
So essentially, they found that agenc coding tasks used a thousand times more tokens than a code chat or just a normal reasoning chat.
00:15:48.399 --> 00:15:54.240
So the same task runs varied up to 30x, making forecasting brutally hard.
00:15:54.480 --> 00:15:55.200
That's the other thing.
00:15:55.279 --> 00:16:03.200
And I think that you know, say what you want about your model of choice, but certain models aren't exactly token efficient, right?
00:16:03.360 --> 00:16:10.159
If you look at the artificial uh artificial analysis leaderboard, the anthropic models are extremely token inefficient.
00:16:10.320 --> 00:16:28.480
Yes, they're some of the best in the world, but if you are not keeping a close eye on your AI spin and you're just giving these very powerful models access to all your data, your repos, your uh code, uh you know, your code base, whatever it is, and you're giving it this big goal and then just not really watching it, right?
00:16:28.799 --> 00:16:40.799
We've seen models like Fable, like Opus be two to three X uh more token inefficient than some of their closest competitors from the other labs, right?
00:16:41.039 --> 00:16:46.799
From um, you know, Google, OpenAI, you know, and then even some of the open source models as well.
00:16:47.039 --> 00:16:58.399
And then there is the rumored story, right, that a single enterprise wasn't keeping a close watch on their API bill and accidentally spent$500 million uh with Anthropic.
00:16:58.480 --> 00:17:05.920
I still don't know if that is true or not, but there were some reputable uh you know news organizations reporting on it.
00:17:06.240 --> 00:17:17.759
But that is the risk, and that is why this essentially AI bill has gone all the way up to the boardroom, and it used to be a couple of months ago, right?
00:17:17.920 --> 00:17:24.400
The um the message being pushed down was let's increase our seats, right?
00:17:24.480 --> 00:17:29.759
Hey, we're paying for you know 5,000 seats, but only 2,500 are being used.
00:17:29.839 --> 00:17:43.039
We need to increase that, and now that has shifted because a lot of those seats, quote unquote, that they were paying for before now have meter billing, maybe on top of a very limited uh, you know, monthly or weekly quota.
00:17:43.200 --> 00:17:54.079
So now, you know, there was all of this push over the end of 2025 and early, early 2026 to let's get people using this AI as much as possible, right?
00:17:54.240 --> 00:17:56.960
We all got this taste of the gateway drugs, right?
00:17:57.119 --> 00:17:59.680
The 2025 models were fantastic.
00:17:59.839 --> 00:18:05.200
The harnessing improved, the tool calling improved, models reasoned and thought by default, right?
00:18:05.279 --> 00:18:16.000
So all of a sudden, you had companies that were smart and doing things the right way, they showed measurable ROI and they're like, let's get this thing, let's use this as much as humanly possible.
00:18:16.160 --> 00:18:26.400
And then the whiplash effect is hitting us all in the face, and that's why we have Javon's paradox, and that explains kind of this concept of cheaper, cheaper tokens but larger bills.
00:18:26.480 --> 00:18:36.640
So if you don't know what that is, that's essentially the idea that as a resource becomes more efficient to use, its consumption often increases rather than decreases.
00:18:36.720 --> 00:18:41.039
So Javon's paradox means cheaper units can drive more total consumption.
00:18:41.200 --> 00:18:41.920
Simple, right?
00:18:42.079 --> 00:18:54.000
But cheaper tokens also created those longer contexts in more agents, more retries, more automation, and sometimes more loops of tool calling in a bad way, right?
00:18:54.160 --> 00:19:00.720
Just models calling tools when you're not looking at them and they're growing over and over and over, and you're maybe not monitoring that.
00:19:00.799 --> 00:19:07.519
And that's why these bills are growing because of demand expanding faster than the unit prices fell.
00:19:07.680 --> 00:19:20.559
And also sometimes the human in the loop being a little lazy and not keeping an eye on because they're like, Hey, I can give this thing a goal and give this all this data, and it's going to work for 16 hours and it's going to get the job done.
00:19:20.720 --> 00:19:30.799
Maybe not looking at how much that job is actually costing or how many you know, times a certain model is trying to go fetch a website and failing and it's just trying over and over and over.
00:19:30.880 --> 00:19:39.200
And maybe each of those tries, maybe there's hundreds or thousands of tries, and it's costing you maybe a couple of dollars every single attempt.
00:19:40.400 --> 00:19:46.240
So there's use cases already we've seen of companies maybe doing this the right way.
00:19:46.319 --> 00:19:48.319
And I think one of them is Coinbase.
00:19:48.480 --> 00:20:01.279
So what they did, and they shared about it a little bit online, is they swapped their default models uh from essentially anthropic to GLM52 and Kimi 2.7.
00:20:01.359 --> 00:20:04.400
And they shared that the combination of that, right?
00:20:04.559 --> 00:20:10.000
Not every single model, uh, right, but a lot of their uh, you know, orchestration.
00:20:10.160 --> 00:20:27.039
So, you know, still handing off uh some of the heavier lifting to maybe an anthropic model, but a lot of that, you know, middle grunt work, the orchestration, the summarization, the writing, some of those things that you maybe are fine with a you know one B tier model.
00:20:27.599 --> 00:20:34.079
Well, they were just going open source with that and using something like ZAI's GLM52 or Kimmy27.
00:20:34.319 --> 00:20:39.200
So combining that with caching, proper caching, and then difficulty-based routing.
00:20:39.359 --> 00:20:44.480
Well, they were able to cut spend to nearly half of what it was at its peak.
00:20:44.640 --> 00:20:47.839
Yet they increased their actual token usage.
00:20:48.000 --> 00:20:52.400
And in terms of output, it didn't fall off, at least according to the company.
00:20:52.720 --> 00:20:59.519
And that's also led to a surge of a new breed of model routers.
00:20:59.680 --> 00:21:00.240
Uh, right.
00:21:00.400 --> 00:21:06.240
And what's funny is well, I told you guys about this more than 18 months ago.
00:21:06.319 --> 00:21:12.240
That's why, uh, especially those, you know, uh year-end uh prediction series.
00:21:12.400 --> 00:21:15.200
That's why you got to listen to these and put these things into practice.
00:21:15.359 --> 00:21:18.000
Because I told you guys this was coming, uh, right?
00:21:18.079 --> 00:21:21.039
The mixture of models and some uh some other things.
00:21:21.200 --> 00:21:31.039
But we've seen a lot of I think great innovations, some of them like open routers fusion, uh perplexity computer, uh computer, although I wish it was a little more affordable.
00:21:31.200 --> 00:21:34.319
Um, you know, and you've seen some other ones as well.
00:21:34.400 --> 00:21:36.960
Uh merge, uh another great example.
00:21:37.039 --> 00:21:42.559
But essentially now you have these services that are kind of plug and play and will do this for you.
00:21:42.720 --> 00:21:47.759
So similarly, how Coinbase went in and did this all manually, right?
00:21:47.920 --> 00:21:54.559
You are having some quote unquote third party services that are going to do a lot of the heavy lifting for you, right?
00:21:54.720 --> 00:21:59.759
You enter in your API keys, uh, you know, connect your data as you normally would.
00:22:00.160 --> 00:22:04.960
Via, you know, easy one click, easy uh, you know, porting over connectors.
00:22:05.119 --> 00:22:09.839
And then, well, you can just hopefully reduce your API spend.
00:22:10.000 --> 00:22:15.599
There are obviously things you have to keep in mind and do your due diligence on in terms of privacy, data sharing, and all of those things.
00:22:15.680 --> 00:22:16.240
Uh, right.
00:22:16.400 --> 00:22:20.319
I'm not gonna be vouching for all of these other companies and how they do that.
00:22:20.480 --> 00:22:24.079
You have to go do your own uh kind of digging on those.
00:22:24.480 --> 00:22:33.759
But routing is just gonna pick the model, but tuning is I think where we're gonna start to see a lot of moats being built.
00:22:33.920 --> 00:22:34.880
Yes, that's right.
00:22:35.119 --> 00:22:36.640
Fine tuning models.
00:22:36.799 --> 00:22:41.279
I don't know if we'd say it's making a comeback because it didn't really go away, right?
00:22:41.359 --> 00:22:53.200
But I think fine tuning was all the rage back in like, you know, late 2023 and 2024, before we had a handful of you know, frontier AI labs to choose from.
00:22:53.440 --> 00:22:57.359
And before, you know, bringing your data was difficult, right?
00:22:57.440 --> 00:23:04.960
Back when you know, we were talking about rag pipelines a lot, also fine-tuning models was a huge deal.
00:23:05.119 --> 00:23:10.319
Um, and I think that again, this is something that we talked about 18 months ago, right?
00:23:10.480 --> 00:23:15.519
Several models kind of teaming up, uh, not just mixture of experts, right?
00:23:15.680 --> 00:23:27.680
Where you're looking at sparse versus dense models uh in giving a query to a big model and a big model just activating the uh, you know, the experts that it needs within those parameters that it carries.
00:23:27.920 --> 00:23:35.759
No, I think we're looking at the mixture of models as being something that is going to become increasingly popular in the near future.
00:23:35.920 --> 00:23:40.480
And I actually hope that we see that from the actual Frontier Labs themselves.
00:23:40.559 --> 00:23:46.640
I would love to see that from uh, you know, Anthropic, OpenAI, Microsoft, Google, etc., right?
00:23:46.720 --> 00:23:54.799
Where even within something like Claude Desktop or uh, you know, codecs where you can go into a certain mode and it will do it for you.
00:23:54.960 --> 00:24:03.759
We've kind of seen previews of this that I don't think were super well done, where you had these auto modes and it decided how much reasoning it would want to use.
00:24:03.920 --> 00:24:09.039
But that was really a little bit more, at least for our non-developers, non-technical people.
00:24:09.200 --> 00:24:18.960
Uh, you know, it it was more of I think just an experiment uh to maybe um, you know, keep compute in check around new model launches.
00:24:19.119 --> 00:24:30.960
But I do think that the combination of mixture of models and fine-tuning models um is going to help companies not just uh keep spend in check, but actually build a moat.
00:24:31.119 --> 00:24:33.759
So this kind of flew under the radar.
00:24:34.000 --> 00:24:42.079
Um, you know, I saw this out when I was out at Microsoft Build Conference, but now Microsoft Foundry offers fine tuning as a service, right?
00:24:42.160 --> 00:24:48.319
So, as an example, you know, Foundry has a ton of models from all of these different providers, right?
00:24:48.480 --> 00:25:02.160
If you just need one model and if it's 80% of your usage, let's just say, as an example, is doing some financial modeling around a certain sector, instead of taking one of the big state-of-the-art trillion parameter models, right?
00:25:02.240 --> 00:25:04.799
You can work with now Microsoft in Foundry.
00:25:05.039 --> 00:25:06.400
They do this as a service.
00:25:06.480 --> 00:25:23.440
Uh, they can find the right, maybe a smaller or medium-sized model and then just fine-tune that or essentially create a version of the of a medium-sized model that's more cost efficient, faster, and well, just better because they're essentially going to train a version of the model just around your specific tasks.
00:25:23.599 --> 00:25:30.160
Speaking of that, that seems to be a uh key area where Thinking Machines Labs is starting to play.
00:25:30.240 --> 00:25:48.400
Uh, right, former uh, you know, OpenAI CTO and co-founder Mira Marathi uh with Thinking Machines Lab, they just came out with a study that showed Bridgewater and Thinking Machines Lab reported a tuned specialist beat a frontier model at about 14x lower cost.
00:25:48.720 --> 00:25:56.240
Uh so that does seem something that Thinking Machines uh is doing as well, essentially offering fine-tuning API as a service.
00:25:56.400 --> 00:26:05.200
But when you have the big labs, right, and sure, we'll throw Thinking Machines in as a big lab in terms of uh market cap valuation money raised, right?
00:26:05.279 --> 00:26:07.119
You they're at least in that consideration.
00:26:07.279 --> 00:26:12.000
Microsoft obviously one of the biggest and best in the country or sorry, in the world, right?
00:26:12.160 --> 00:26:34.160
When you are starting to see fine tuning as a service, that is not um about model specialization, that is about the uh the impending uh reality of subscription models going away in enterprise companies having to look at that, right?
00:26:34.240 --> 00:26:44.000
If you were paying for, as an example, if you were paying for 10,000 licenses of Microsoft GitHub Copilot, and then all of a sudden they switched over to us at billing, right?
00:26:44.160 --> 00:26:45.759
Are you gonna continue to do that?
00:26:46.000 --> 00:26:46.960
I don't know, right?
00:26:47.039 --> 00:26:49.839
This is where you're gonna have to start looking at these alternatives.
00:26:50.160 --> 00:26:52.720
So here is your Monday move.
00:26:52.960 --> 00:26:55.920
You have to build the spend router, right?
00:26:56.000 --> 00:27:07.759
Whether you look at some of those other services, uh, whether you start doing this and duct taping it yourself internally, or whether you are just preparing for when the big labs offer this, you have to start putting this in place.
00:27:07.920 --> 00:27:10.079
You don't just cut AI blindly.
00:27:10.160 --> 00:27:12.000
That is the absolute worst thing you can do.
00:27:12.160 --> 00:27:15.599
And unfortunately, a lot of companies are doing this.
00:27:15.759 --> 00:27:17.119
So don't do that.
00:27:17.279 --> 00:27:17.759
Uh, right.
00:27:17.839 --> 00:27:24.240
You have to start classifying work by value, risk, and complexity and start using some of these cheaper open source models.
00:27:24.319 --> 00:27:29.519
Again, go through all your due diligence on the privacy data, uh, you know, IP side, all that.
00:27:29.599 --> 00:27:30.079
Uh, right.
00:27:30.160 --> 00:27:45.279
But you should be looking at these tuned specialists, routers, caps, and frontier escalation because the winners ultimately are going to be the ones, I think, like Coinbase that are finding ways to actually spend more tokens and to use more tokens.
00:27:45.359 --> 00:27:56.160
I don't think there was anything wrong per se with the token maxing era, but it was more of doing it irresponsibly on the API side and using the most expensive models, right?
00:27:56.240 --> 00:28:09.680
I do think that you have to have some experimentation on these models that are a fraction of the cost, and you still get maybe, you know, 85 to 95 percent of the performance of the state-of-the-art models.
00:28:10.079 --> 00:28:17.359
So here is your seven-step playbook uh to control runaway AI spend as we wrap today's show.
00:28:17.519 --> 00:28:19.519
Number one, you have to meter first.
00:28:19.680 --> 00:28:24.000
So you have to get token spend visibility by team, task, and model, right?
00:28:24.079 --> 00:28:29.279
So, similarly, how I said, hey, right now I'm going through personally 2.5 billion tokens a week.
00:28:29.440 --> 00:28:30.880
I can break that down by project.
00:28:30.960 --> 00:28:34.000
I actually build a tool inside of Codex that helps me do that, right?
00:28:34.160 --> 00:28:35.680
You can be doing these things as well.
00:28:35.839 --> 00:28:44.000
You have to start seeing where your tokens are going, who is spending them, uh, what task, what model, and what are the outcomes as well.
00:28:44.079 --> 00:28:45.279
Number two, set budgets.
00:28:45.440 --> 00:28:51.119
After you do that, you have to be able to cap spend per team, user, and workflow with overage alerts.
00:28:51.279 --> 00:28:54.319
That doesn't mean you cut use, right?
00:28:54.480 --> 00:28:59.279
So let's say you first meter and you say, here's what we're doing on our subscriptions.
00:28:59.599 --> 00:29:05.279
What is that uh, what is that spend that you're gonna have to do to get the same outcomes, right?
00:29:05.440 --> 00:29:10.880
So don't cut budget without protecting your outcomes, all right.
00:29:10.960 --> 00:29:14.079
Number three, you have to start learning to default to cheap, right?
00:29:14.160 --> 00:29:28.160
I'm not saying the cheapest models, but now we do have models that are extremely capable, open weight models like GLM 5.2, uh, you know, like uh Kimmy um 2.7, right?
00:29:28.319 --> 00:29:30.960
We're gonna, I'm sure, have some other things from Quen.
00:29:31.200 --> 00:29:32.400
Uh, they have some great models.
00:29:32.559 --> 00:29:34.799
Deep Seek, I'm sure we'll come up with another good one.
00:29:34.960 --> 00:29:45.519
Again, go through all your proper uh IP privacy security, but you have to start looking at these maybe open source, open weight models, whether you're running them locally or paying for them on the API side.
00:29:45.599 --> 00:29:50.000
They're normally a fraction of the price as the closed uh frontier models.
00:29:50.160 --> 00:29:52.319
Number four, you have to route by difficulty.
00:29:52.480 --> 00:30:00.319
You shouldn't be, as an example, having Fable five on extra ultra reasoning, whatever it's called inside Cloud Code.
00:30:00.400 --> 00:30:03.680
You shouldn't be doing that to help you write better emails, right?
00:30:03.839 --> 00:30:08.079
You have to send the right type of tasks to the right models.
00:30:08.240 --> 00:30:12.480
Uh, then you need to understand your get proper uh caching.
00:30:12.640 --> 00:30:18.559
You need to trim and reuse repeated prompts and start lean sessions per task, right?
00:30:18.640 --> 00:30:24.079
Even simple things like if you're working on the front end using things like projects, right?
00:30:24.160 --> 00:30:29.759
Uh certain uh, you know, certain providers do things better than others, right?
00:30:29.839 --> 00:30:34.319
As an example, Codex is great at compaction, auto compaction, it's really good.
00:30:34.480 --> 00:30:45.680
Um, you know, Claude recommends as chats get longer to start new chats, uh, to save on cost, but you have to start doing those kind of non-technical best practices as well as caching and trimming.
00:30:45.839 --> 00:30:48.000
Number six, fine-tune the repeats.
00:30:48.160 --> 00:30:57.039
Yes, if especially I think if you are a uh Microsoft organization, I do assume that we're gonna start seeing fine-tuning as a service, right?
00:30:57.200 --> 00:31:07.839
I think Thinking Machine Labs and Microsoft uh offering this leads me to believe we're gonna see this probably within uh nine months from all of the big players.
00:31:08.000 --> 00:31:13.200
You have to start looking at and understanding what it's what are we gonna fine tune, right?
00:31:13.279 --> 00:31:19.200
What type of model is going to do 80% of our AI use for certain teams?
00:31:19.359 --> 00:31:27.680
And well, if you fine-tune a model, it might be a little bit of work and a little bit of cost up front, but you have to be able to math the math because chances are it's gonna save you a lot in the long run.
00:31:27.839 --> 00:31:31.440
And then last but not least, escalate on purpose.
00:31:31.759 --> 00:31:40.319
Call the frontier model only to close hard problems, especially if that frontier model is not included in a subscription.
00:31:40.400 --> 00:31:43.599
And I say that as a very timely reminder.
00:31:43.759 --> 00:31:48.160
Yes, Fables, uh Fable Five, a couple more hours, and it's gone.
00:31:48.319 --> 00:31:57.039
We'll see what happens with open AI, uh, GPT-5.6, Soul, uh, and Soul Ultra, they did say it's gonna be included uh inside codex.
00:31:57.279 --> 00:31:59.200
So who knows what's happening?
00:31:59.279 --> 00:32:06.880
But all I know right now, as we wrap up today's show, the future of AI is token efficiency.
00:32:07.039 --> 00:32:19.839
And it is no longer uh, if you are the decision maker, six months ago, regardless of what tool, what system, the right move was to use as much AI as possible.
00:32:20.000 --> 00:32:34.960
And that still may be the case, especially if you're an open AI organization, because they are pretty much the only player right now that hasn't drastically uh either cut uh the subscription limits or that hasn't gone to true metered access.
00:32:35.119 --> 00:32:43.920
But chances are some of your team, some of your organizations most needed, most used AI tools have converted to metered pricing.
00:32:44.160 --> 00:32:46.160
So that doesn't mean stop.
00:32:46.720 --> 00:32:48.960
That just means be smarter.
00:32:49.119 --> 00:32:54.400
Don't stop experimenting and well, follow us daily because we're gonna be helping you every step along the way.
00:32:54.559 --> 00:32:56.240
I hope this one was helpful.
00:32:56.400 --> 00:33:09.119
If so, please make sure to go to starthirseries.com, go sign up for our community, uh, go network with other people who are leaders in the AI industry, uh, figuring it all out there.
00:33:09.200 --> 00:33:16.240
Uh also, if this was helpful, please subscribe to the show on Spotify and on Apple Podcasts.
00:33:16.319 --> 00:33:17.119
Thank you for tuning in.
00:33:17.279 --> 00:33:20.480
Hope to see you back tomorrow and every day for more everyday AI.
00:33:20.559 --> 00:33:21.519
Thanks, y'all.
00:00:01.600 --> 00:00:11.359
For the past three and a half years, AI has almost felt too cheap to be real, especially for power users during early 2026.
00:00:11.679 --> 00:00:27.760
I mean, you pay a monthly fee, you open Chat GPT or Claude or Codex, Gemini Copilot, whatever, and then you have agents run for hours completing your actual work with your contacts without paying an additional penny on top of that subscription.
00:00:28.079 --> 00:00:33.840
But that version of AI is ending and it's gonna hit even harder later today.
00:00:34.000 --> 00:00:34.479
Why?
00:00:34.719 --> 00:00:43.759
Well, that's because Anthropic's Fable 5 only has hours left until it's no longer included in paid subscription plans.
00:00:44.000 --> 00:00:46.960
And that's not just an anthropic story.
00:00:47.119 --> 00:00:52.079
And it's the clearest signal yet that Frontier AI is becoming a metered utility.
00:00:52.240 --> 00:00:56.240
I mean, Microsoft GitHub moved co-pilot into AI credits.
00:00:56.399 --> 00:01:14.719
Google changed Gemini limits around compute complexity features and chat length, and Microsoft started pricing co-pilot cowork through credits, not included subscriptions, and even NeoClaude XAI's Grok has been moving users toward share weekly pools and extra usage credits.
00:01:14.959 --> 00:01:25.920
So with the explosion of agentic models that can now work for hours with expert level artifacts, the question is no longer how do we get everyone in our company using AI?
00:01:26.079 --> 00:01:32.159
The board level question is how do we use more AI without bankrupting our budget in the process?
00:01:32.319 --> 00:01:35.200
And that's exactly what we're going to be tackling today.
00:01:35.439 --> 00:01:37.359
So here is the big picture.
00:01:37.599 --> 00:01:41.359
Frontier AI has definitely now become a metered utility.
00:01:41.599 --> 00:01:54.640
And the wild west days of using whatever model you wanted around the clock and just paying a$20 or$200 a month subscription fee and nothing more, those days are pretty much gone.
00:01:54.879 --> 00:02:01.519
Um like I said, for three years we've been doing this seemingly unlimited use of AI.
00:02:01.599 --> 00:02:13.120
And if you go back to literally back in 2023, even when the models were pre-agentic, I even said eventually we're gonna see 200,$2,000 subscription plans.
00:02:13.199 --> 00:02:15.599
And I think even at that time, it's still gonna be a steal.
00:02:15.840 --> 00:02:28.479
And now when I throw some numbers out, you're gonna see that even these$200 a month plans, or if there isn't even a plan on top of that someday, will still be a steal considering what you're gonna start paying tomorrow.
00:02:28.639 --> 00:02:44.159
That's because even though the price per token has fallen by literally like 98% over the past three years, depending on what tier you're looking at, the agentic capabilities and the amount of tokens that these models are using has legit skyrocketed.
00:02:44.319 --> 00:02:53.360
And over the past two or three months, AI has quickly gone from let's use as much as we can in token max to putting strict spending limits.
00:02:53.520 --> 00:03:03.520
And sometimes, you know, developers at huge companies aren't even to use AI as much as everyday non-technical people like you and me.
00:03:03.599 --> 00:03:14.000
And I think the fable issue today in Thropix Fable V leaving subscription uh in a couple of hours only heightens that chatbot bill problem.
00:03:14.159 --> 00:03:16.319
So on today's show, here's what you're gonna learn.
00:03:16.400 --> 00:03:25.360
You're gonna learn why Fable Five subscription exit is the cleanest and loudest warning shot for subscription AI and your team's AI strategy.
00:03:25.439 --> 00:03:31.439
You're gonna know why GitHub, Copilot, and Grok swapped unlimited access for pay-per-use meters.
00:03:31.520 --> 00:03:37.680
You're gonna understand why companies are capping spend, but still trying to use more tokens and how they're doing it.
00:03:37.840 --> 00:03:44.000
And I'm gonna leave you at the very end with our seven-step playbook to control runaway AI spend.
00:03:44.159 --> 00:03:45.199
Let's get into it.
00:03:45.439 --> 00:03:48.960
Welcome to Everyday AI and our Start Here series.
00:03:49.120 --> 00:03:55.280
This is the essential podcast series to both learn the AI basics and to double down on your AI knowledge.
00:03:55.520 --> 00:04:01.840
So if this is helpful and if that's exactly what you're trying to do, make sure you go to starthierseries.com.
00:04:01.919 --> 00:04:06.000
That is going to give you exclusive access to our inner circle community.
00:04:06.159 --> 00:04:20.079
You're gonna be thrust straight into our Start Here Series space where you can go listen uh to every single Start Here series, the simplest way to do it on the playlist that we have there, as well as you can go read about it, connect with other people who are trying to do the same thing.
00:04:20.319 --> 00:04:33.040
And if you missed our last Start Here series episode, we talked about the desktop agent lingo simplified, going over goals, loops, plans, sub agents, and how it works in codex and clawed code.
00:04:33.120 --> 00:04:49.120
And well, if you're doing all of those things, chances are yeah, your bills might start getting more and more expensive, which brings us to today's episode AI cost control 101 and why your chat bot bill is now becoming a board level problem.
00:04:49.759 --> 00:04:56.639
So over the past couple of months, we've seen this, and it's not just Anthropic's Fable Five.
00:04:56.800 --> 00:04:59.920
I mean, let's go back to May, right after Google's I.O.
00:05:00.079 --> 00:05:11.040
conference, their big yearly conference, for the first time, Google started putting actually pretty strict restrictions uh on their Google Gemini.
00:05:11.279 --> 00:05:18.160
So it's actually the first time I'd ever uh run into Google Gemini limits on their$20 plan.
00:05:18.319 --> 00:05:20.639
Up until then, it had been pretty generous.
00:05:20.800 --> 00:05:36.319
And, you know, after they made this shift of actually starting to count longer, uh, you know, taking things like complexity, the model you use, and chat length, all of a sudden that$20 plan that used to feel pretty great started feeling not super useful.
00:05:36.480 --> 00:05:45.279
Uh, then in June, you had Microsoft's GitHub Copilot replaced their premium request with token-based AI credits.
00:05:45.439 --> 00:05:49.759
So the unlimited plan was gone for Microsoft's GitHub Copilot.
00:05:49.920 --> 00:05:52.079
The same thing with Copilot Cowork.
00:05:52.240 --> 00:06:02.639
Microsoft is starting to go with task-level credit pricing and no longer including essentially unlimited co-work usage after they moved it to general availability.
00:06:02.720 --> 00:06:12.639
Uh, then a couple days ago, we also saw Grok uh from XAI Space XAI, I think is technically their new name, as of a couple hours ago.
00:06:12.720 --> 00:06:22.240
Uh, right, they started rolling out uh the credit base system as well, which is pretty telling considering, well, they're kind of like a neo-cloud leader now, right?
00:06:22.319 --> 00:06:28.800
They are uh one of the companies that has technically extra compute and they're even charging for it.
00:06:28.959 --> 00:06:39.279
And like I've already said, today is maybe the day that all of this gets thrust more squarely into focus because of Anthropic's Fable 5.
00:06:39.439 --> 00:06:41.120
And why is that important?
00:06:41.360 --> 00:07:01.040
Well, right now, at least, I mean, we'll see what happens over the next couple of hours because we've heard uh rumors that we might get a GPT-5-6 from OpenAI today, uh, sometime between today and Thursday, but uh, at least for the next couple of hours, uh, Fable Five is the most powerful AI model in the world.
00:07:01.199 --> 00:07:05.920
And at least for a couple more hours, it is still available in a subscription.
00:07:06.000 --> 00:07:10.720
But after today, you are going to be paying the full API price for that.
00:07:11.120 --> 00:07:20.800
So we've gone over the uh Fable Five drama kind of enough, but you know, they put it out there, they had to take it away because of some problems with the US government.
00:07:20.959 --> 00:07:27.360
Uh, and they brought it back a couple of days ago, but they said only included in plans through July 7th.
00:07:27.439 --> 00:07:32.000
And they said, hey, maybe one day when we have enough compute, maybe we'll bring it back, but no guarantee.
00:07:32.160 --> 00:07:33.600
And why does this matter?
00:07:33.839 --> 00:07:43.199
Because, like I said, for the past three and a half years, we've been able to, you know, whether you have a$20 plan,$100 pro max plan,$200 pro max, right?
00:07:43.360 --> 00:07:48.000
But for a very low price, and I know that's all relative, right?
00:07:48.160 --> 00:07:55.040
For many people out there, I understand a$200 a month plan, especially if you're paying it for it yourself, doesn't seem like the cheapest thing.
00:07:55.199 --> 00:07:57.519
But let me just show you why it actually is.
00:07:57.839 --> 00:08:02.560
So now, starting tomorrow, uh technically 11:59 p.m.
00:08:02.879 --> 00:08:12.079
Pacific time, you're gonna be paying uh$10 per million input tokens and$50 per million output tokens for uh Fable 5.
00:08:12.560 --> 00:08:14.800
Let me just put this into comparison now.
00:08:14.959 --> 00:08:17.360
So you're probably saying, all right, what's the big deal?
00:08:17.519 --> 00:08:31.439
All right, like Jordan, if you're paying, you know,$200 uh, you know, a month for, you know, which I am, I'm paying$200 a month uh for the Claude 20X Max plan, I'm paying$200 a month for the ChatGPT Pro plan.
00:08:31.680 --> 00:08:37.360
Uh, you know, I paid the$200 a month previously for the Google uh Ultra plan, so I paid for all of these.
00:08:37.519 --> 00:08:39.360
So you're saying, okay, what's the big deal?
00:08:39.519 --> 00:08:44.480
You know, if it just moves to API only, can't you just use that money for the API credits?
00:08:44.799 --> 00:08:45.679
Yeah, no.
00:08:46.000 --> 00:08:46.320
Okay.
00:08:46.799 --> 00:08:53.519
So granted, OpenAI has still ridiculously generous limits.
00:08:53.679 --> 00:08:54.000
All right.
00:08:54.159 --> 00:08:55.519
So let me put that out there.
00:08:55.679 --> 00:09:12.000
You're normally not getting that level of spend on these other plans, but I've been averaging uh over the past couple of weeks about two-ish billion tokens a week on the$200 a month uh OpenAI codex plan.
00:09:12.320 --> 00:09:15.279
Okay, uh, which is included with ChatGPT, right?
00:09:15.360 --> 00:09:16.159
So it's the same plan.
00:09:16.480 --> 00:09:21.200
Uh so that at a at a very generous blended rate, right?
00:09:21.279 --> 00:09:32.639
Of uh a blended rate of input and output tokens, that is a$200,000 a month anthropic bill.
00:09:32.879 --> 00:09:48.799
If I were to take everything I'm doing inside of Codex and I'm only paying$200 a month, and if I were to say, in theory, move all of that into um Fable, that is$200,000 a month.
00:09:49.840 --> 00:10:01.679
Okay, now you see why this has been a such a big discussion in the AI world about no longer having this democratized AI uh access to AI.
00:10:01.840 --> 00:10:06.399
We'll see uh, you know, what it means with uh OpenAI's models moving forward.
00:10:06.480 --> 00:10:18.799
Uh, they did say that their, you know, new GPT-5.6, uh kind of the different tiers, uh, Seoul, Terra, Luna, and you know, Soul Ultra would be included in subscription plans.
00:10:18.960 --> 00:10:35.360
So maybe OpenAI may be one of the only few players left that has yet to um, you know, kind of knock, knock out that subsidy, but that is still the reality uh because you have to assume that one day that's gonna be the status quo.
00:10:35.440 --> 00:10:41.360
And maybe open AI might be able to do it for another three months, three quarters, three years.
00:10:41.519 --> 00:10:42.399
We're not sure.
00:10:42.639 --> 00:10:52.159
But eventually, AI is going to get more and more expensive because the models themselves, in theory, are going to be eating up more and more tokens.
00:10:52.399 --> 00:10:57.200
And this is exactly why we're seeing that token maxing to token efficiency shift.
00:10:57.279 --> 00:11:02.080
And we did cover that on the start here series, volume 27 or episode 789.
00:11:02.159 --> 00:11:03.200
So go check that out.
00:11:03.279 --> 00:11:05.919
But we've seen it in some of the biggest companies in the world.
00:11:06.159 --> 00:11:16.879
So, as an example, uh, we saw the very uh you know popular story, Uber reportedly burned through its 2026 AI coding budget in just four months.
00:11:17.120 --> 00:11:22.799
Tesla, we recently saw capped employee AI tool spend to just$200 a week.
00:11:22.960 --> 00:11:30.559
And then UBS says that 60% of interviewed enterprises are throttling AI spent already, right?
00:11:30.720 --> 00:11:35.519
So I do assume that number will probably go up if UBS does the same survey quarterly.
00:11:35.600 --> 00:11:36.480
I don't know if they do.
00:11:36.720 --> 00:11:51.679
I would assume that's gonna go up to about 70 to 80 percent uh by the end of the third quarter, because I don't think the uh kind of the token efficiency uh wall has hit the majority of enterprises just yet.
00:11:52.000 --> 00:11:56.000
So, and this is actually everywhere, right?
00:11:56.080 --> 00:12:06.720
I even had a conversation with uh a friend who is in software development at one of the Mag 7 companies, you know, and he said his spend is$100 a month, right?
00:12:06.799 --> 00:12:09.279
So that$200 a week at Tesla.
00:12:09.440 --> 00:12:12.320
Although when you're looking at API credits, you're like, that's like nothing.
00:12:12.480 --> 00:12:15.919
Yeah, imagine having like$100 a month.
00:12:16.000 --> 00:12:17.120
So that is the reality.
00:12:17.279 --> 00:12:27.440
So I also want to tell this to people out there who maybe don't talk to a lot of people at Fortune 100 type companies and you know, big mag 7 type companies.
00:12:27.519 --> 00:12:32.480
We are seeing actual real spend limits on the amount of AI that these employees can use.
00:12:32.639 --> 00:12:40.799
And that actually, I think, for SMBs and medium-sized enterprises might create a strategic advantage, right?
00:12:40.879 --> 00:12:50.799
If the biggest players are maybe uh, you know, having to reel in their spend, and maybe part of that was their own fault for pushing these token leaderboards and all these other things.
00:12:50.960 --> 00:12:54.960
I think there is an advantage to be had for everyone else.
00:12:55.279 --> 00:12:59.759
All right, let's take a quick break for a word from our partners.
00:13:00.240 --> 00:13:05.279
Here's what most AI tools still can't do: work outside their own little box.
00:13:05.440 --> 00:13:07.759
The all new Slack bot just changed that.
00:13:07.840 --> 00:13:14.720
It's your AI teammate inside Slack, and now it can read, write, and act across the other apps your team already uses.
00:13:14.879 --> 00:13:20.240
No more tool switching, no more re-explaining yourself every time you open a new tab.
00:13:20.399 --> 00:13:26.879
One ops team at engine says the summary feature alone saves them 15 to 20 minutes of use.
00:13:27.039 --> 00:13:30.320
See what Slackbot can do at Slack.com.
00:13:32.720 --> 00:13:34.799
Okay, so why now?
00:13:35.279 --> 00:13:35.600
Right?
00:13:35.759 --> 00:13:42.080
Why are we seeing this uh, you know, your chatbot bill becoming a boardroom problem now?
00:13:42.240 --> 00:13:44.000
It sure wasn't at this time last year.
00:13:44.080 --> 00:13:46.080
It wasn't really at the end of 2025.
00:13:46.240 --> 00:13:55.440
And I think to make a very short story of the last six to eight months uh of AI developments, if you listen to the show, it's no surprise.
00:13:55.600 --> 00:13:57.519
Maybe if you're newer here, let me fill you in.
00:13:57.679 --> 00:14:06.559
And Tropic set the pace at the end of 2025 uh with Claude Code, bringing that to the desktop uh as well as Claude Co-work in early 2026.
00:14:06.639 --> 00:14:12.240
And then I think the rest of the way has been blazed uh by Codecs, Cursor, and some others.
00:14:12.480 --> 00:14:18.159
But we are now having extremely powerful AI models, even for non-technical work, right?
00:14:18.240 --> 00:14:24.480
When I talked about spending, you know, two-ish billion tokens a week, I'm not doing a ton of traditional coding.
00:14:24.559 --> 00:14:33.440
Yes, I'm building a bunch of programs that I use all the time, but a lot of my token spend is just doing right scheduled work that I wouldn't normally be doing, right?
00:14:33.519 --> 00:14:36.720
Doing a bunch of research, creating a bunch of just documents.
00:14:36.879 --> 00:14:46.080
Um, and the reason why we are seeing this explosion is because these capabilities are becoming more powerful and they're becoming easier for non-technical people to use.
00:14:46.240 --> 00:14:55.120
So you have more and more people starting to automate their everyday work with these desktop agents that can work 24-7 and can work for hours, right?
00:14:55.279 --> 00:15:04.159
I remember even in the earlier days in 2026 with Claude Code, I would have to really push the models uh to work for maybe like an hour or so.
00:15:04.320 --> 00:15:20.000
But now with the combination of subagents and plan mode and goal and all those things that we talked about in the last start here series, every single day I'm usually having multiple threads running for eight to 16 hours and creating amazing outcomes.
00:15:20.320 --> 00:15:24.960
So it's all about the amount of tokens being spent.
00:15:25.200 --> 00:15:37.440
So a Stanford Digital Economy Lab study this year showed how much more impactful and how much more token hungry today's models are than the old chatbot only non-reasoning models.
00:15:37.600 --> 00:15:48.320
So essentially, they found that agenc coding tasks used a thousand times more tokens than a code chat or just a normal reasoning chat.
00:15:48.399 --> 00:15:54.240
So the same task runs varied up to 30x, making forecasting brutally hard.
00:15:54.480 --> 00:15:55.200
That's the other thing.
00:15:55.279 --> 00:16:03.200
And I think that you know, say what you want about your model of choice, but certain models aren't exactly token efficient, right?
00:16:03.360 --> 00:16:10.159
If you look at the artificial uh artificial analysis leaderboard, the anthropic models are extremely token inefficient.
00:16:10.320 --> 00:16:28.480
Yes, they're some of the best in the world, but if you are not keeping a close eye on your AI spin and you're just giving these very powerful models access to all your data, your repos, your uh code, uh you know, your code base, whatever it is, and you're giving it this big goal and then just not really watching it, right?
00:16:28.799 --> 00:16:40.799
We've seen models like Fable, like Opus be two to three X uh more token inefficient than some of their closest competitors from the other labs, right?
00:16:41.039 --> 00:16:46.799
From um, you know, Google, OpenAI, you know, and then even some of the open source models as well.
00:16:47.039 --> 00:16:58.399
And then there is the rumored story, right, that a single enterprise wasn't keeping a close watch on their API bill and accidentally spent$500 million uh with Anthropic.
00:16:58.480 --> 00:17:05.920
I still don't know if that is true or not, but there were some reputable uh you know news organizations reporting on it.
00:17:06.240 --> 00:17:17.759
But that is the risk, and that is why this essentially AI bill has gone all the way up to the boardroom, and it used to be a couple of months ago, right?
00:17:17.920 --> 00:17:24.400
The um the message being pushed down was let's increase our seats, right?
00:17:24.480 --> 00:17:29.759
Hey, we're paying for you know 5,000 seats, but only 2,500 are being used.
00:17:29.839 --> 00:17:43.039
We need to increase that, and now that has shifted because a lot of those seats, quote unquote, that they were paying for before now have meter billing, maybe on top of a very limited uh, you know, monthly or weekly quota.
00:17:43.200 --> 00:17:54.079
So now, you know, there was all of this push over the end of 2025 and early, early 2026 to let's get people using this AI as much as possible, right?
00:17:54.240 --> 00:17:56.960
We all got this taste of the gateway drugs, right?
00:17:57.119 --> 00:17:59.680
The 2025 models were fantastic.
00:17:59.839 --> 00:18:05.200
The harnessing improved, the tool calling improved, models reasoned and thought by default, right?
00:18:05.279 --> 00:18:16.000
So all of a sudden, you had companies that were smart and doing things the right way, they showed measurable ROI and they're like, let's get this thing, let's use this as much as humanly possible.
00:18:16.160 --> 00:18:26.400
And then the whiplash effect is hitting us all in the face, and that's why we have Javon's paradox, and that explains kind of this concept of cheaper, cheaper tokens but larger bills.
00:18:26.480 --> 00:18:36.640
So if you don't know what that is, that's essentially the idea that as a resource becomes more efficient to use, its consumption often increases rather than decreases.
00:18:36.720 --> 00:18:41.039
So Javon's paradox means cheaper units can drive more total consumption.
00:18:41.200 --> 00:18:41.920
Simple, right?
00:18:42.079 --> 00:18:54.000
But cheaper tokens also created those longer contexts in more agents, more retries, more automation, and sometimes more loops of tool calling in a bad way, right?
00:18:54.160 --> 00:19:00.720
Just models calling tools when you're not looking at them and they're growing over and over and over, and you're maybe not monitoring that.
00:19:00.799 --> 00:19:07.519
And that's why these bills are growing because of demand expanding faster than the unit prices fell.
00:19:07.680 --> 00:19:20.559
And also sometimes the human in the loop being a little lazy and not keeping an eye on because they're like, Hey, I can give this thing a goal and give this all this data, and it's going to work for 16 hours and it's going to get the job done.
00:19:20.720 --> 00:19:30.799
Maybe not looking at how much that job is actually costing or how many you know, times a certain model is trying to go fetch a website and failing and it's just trying over and over and over.
00:19:30.880 --> 00:19:39.200
And maybe each of those tries, maybe there's hundreds or thousands of tries, and it's costing you maybe a couple of dollars every single attempt.
00:19:40.400 --> 00:19:46.240
So there's use cases already we've seen of companies maybe doing this the right way.
00:19:46.319 --> 00:19:48.319
And I think one of them is Coinbase.
00:19:48.480 --> 00:20:01.279
So what they did, and they shared about it a little bit online, is they swapped their default models uh from essentially anthropic to GLM52 and Kimi 2.7.
00:20:01.359 --> 00:20:04.400
And they shared that the combination of that, right?
00:20:04.559 --> 00:20:10.000
Not every single model, uh, right, but a lot of their uh, you know, orchestration.
00:20:10.160 --> 00:20:27.039
So, you know, still handing off uh some of the heavier lifting to maybe an anthropic model, but a lot of that, you know, middle grunt work, the orchestration, the summarization, the writing, some of those things that you maybe are fine with a you know one B tier model.
00:20:27.599 --> 00:20:34.079
Well, they were just going open source with that and using something like ZAI's GLM52 or Kimmy27.
00:20:34.319 --> 00:20:39.200
So combining that with caching, proper caching, and then difficulty-based routing.
00:20:39.359 --> 00:20:44.480
Well, they were able to cut spend to nearly half of what it was at its peak.
00:20:44.640 --> 00:20:47.839
Yet they increased their actual token usage.
00:20:48.000 --> 00:20:52.400
And in terms of output, it didn't fall off, at least according to the company.
00:20:52.720 --> 00:20:59.519
And that's also led to a surge of a new breed of model routers.
00:20:59.680 --> 00:21:00.240
Uh, right.
00:21:00.400 --> 00:21:06.240
And what's funny is well, I told you guys about this more than 18 months ago.
00:21:06.319 --> 00:21:12.240
That's why, uh, especially those, you know, uh year-end uh prediction series.
00:21:12.400 --> 00:21:15.200
That's why you got to listen to these and put these things into practice.
00:21:15.359 --> 00:21:18.000
Because I told you guys this was coming, uh, right?
00:21:18.079 --> 00:21:21.039
The mixture of models and some uh some other things.
00:21:21.200 --> 00:21:31.039
But we've seen a lot of I think great innovations, some of them like open routers fusion, uh perplexity computer, uh computer, although I wish it was a little more affordable.
00:21:31.200 --> 00:21:34.319
Um, you know, and you've seen some other ones as well.
00:21:34.400 --> 00:21:36.960
Uh merge, uh another great example.
00:21:37.039 --> 00:21:42.559
But essentially now you have these services that are kind of plug and play and will do this for you.
00:21:42.720 --> 00:21:47.759
So similarly, how Coinbase went in and did this all manually, right?
00:21:47.920 --> 00:21:54.559
You are having some quote unquote third party services that are going to do a lot of the heavy lifting for you, right?
00:21:54.720 --> 00:21:59.759
You enter in your API keys, uh, you know, connect your data as you normally would.
00:22:00.160 --> 00:22:04.960
Via, you know, easy one click, easy uh, you know, porting over connectors.
00:22:05.119 --> 00:22:09.839
And then, well, you can just hopefully reduce your API spend.
00:22:10.000 --> 00:22:15.599
There are obviously things you have to keep in mind and do your due diligence on in terms of privacy, data sharing, and all of those things.
00:22:15.680 --> 00:22:16.240
Uh, right.
00:22:16.400 --> 00:22:20.319
I'm not gonna be vouching for all of these other companies and how they do that.
00:22:20.480 --> 00:22:24.079
You have to go do your own uh kind of digging on those.
00:22:24.480 --> 00:22:33.759
But routing is just gonna pick the model, but tuning is I think where we're gonna start to see a lot of moats being built.
00:22:33.920 --> 00:22:34.880
Yes, that's right.
00:22:35.119 --> 00:22:36.640
Fine tuning models.
00:22:36.799 --> 00:22:41.279
I don't know if we'd say it's making a comeback because it didn't really go away, right?
00:22:41.359 --> 00:22:53.200
But I think fine tuning was all the rage back in like, you know, late 2023 and 2024, before we had a handful of you know, frontier AI labs to choose from.
00:22:53.440 --> 00:22:57.359
And before, you know, bringing your data was difficult, right?
00:22:57.440 --> 00:23:04.960
Back when you know, we were talking about rag pipelines a lot, also fine-tuning models was a huge deal.
00:23:05.119 --> 00:23:10.319
Um, and I think that again, this is something that we talked about 18 months ago, right?
00:23:10.480 --> 00:23:15.519
Several models kind of teaming up, uh, not just mixture of experts, right?
00:23:15.680 --> 00:23:27.680
Where you're looking at sparse versus dense models uh in giving a query to a big model and a big model just activating the uh, you know, the experts that it needs within those parameters that it carries.
00:23:27.920 --> 00:23:35.759
No, I think we're looking at the mixture of models as being something that is going to become increasingly popular in the near future.
00:23:35.920 --> 00:23:40.480
And I actually hope that we see that from the actual Frontier Labs themselves.
00:23:40.559 --> 00:23:46.640
I would love to see that from uh, you know, Anthropic, OpenAI, Microsoft, Google, etc., right?
00:23:46.720 --> 00:23:54.799
Where even within something like Claude Desktop or uh, you know, codecs where you can go into a certain mode and it will do it for you.
00:23:54.960 --> 00:24:03.759
We've kind of seen previews of this that I don't think were super well done, where you had these auto modes and it decided how much reasoning it would want to use.
00:24:03.920 --> 00:24:09.039
But that was really a little bit more, at least for our non-developers, non-technical people.
00:24:09.200 --> 00:24:18.960
Uh, you know, it it was more of I think just an experiment uh to maybe um, you know, keep compute in check around new model launches.
00:24:19.119 --> 00:24:30.960
But I do think that the combination of mixture of models and fine-tuning models um is going to help companies not just uh keep spend in check, but actually build a moat.
00:24:31.119 --> 00:24:33.759
So this kind of flew under the radar.
00:24:34.000 --> 00:24:42.079
Um, you know, I saw this out when I was out at Microsoft Build Conference, but now Microsoft Foundry offers fine tuning as a service, right?
00:24:42.160 --> 00:24:48.319
So, as an example, you know, Foundry has a ton of models from all of these different providers, right?
00:24:48.480 --> 00:25:02.160
If you just need one model and if it's 80% of your usage, let's just say, as an example, is doing some financial modeling around a certain sector, instead of taking one of the big state-of-the-art trillion parameter models, right?
00:25:02.240 --> 00:25:04.799
You can work with now Microsoft in Foundry.
00:25:05.039 --> 00:25:06.400
They do this as a service.
00:25:06.480 --> 00:25:23.440
Uh, they can find the right, maybe a smaller or medium-sized model and then just fine-tune that or essentially create a version of the of a medium-sized model that's more cost efficient, faster, and well, just better because they're essentially going to train a version of the model just around your specific tasks.
00:25:23.599 --> 00:25:30.160
Speaking of that, that seems to be a uh key area where Thinking Machines Labs is starting to play.
00:25:30.240 --> 00:25:48.400
Uh, right, former uh, you know, OpenAI CTO and co-founder Mira Marathi uh with Thinking Machines Lab, they just came out with a study that showed Bridgewater and Thinking Machines Lab reported a tuned specialist beat a frontier model at about 14x lower cost.
00:25:48.720 --> 00:25:56.240
Uh so that does seem something that Thinking Machines uh is doing as well, essentially offering fine-tuning API as a service.
00:25:56.400 --> 00:26:05.200
But when you have the big labs, right, and sure, we'll throw Thinking Machines in as a big lab in terms of uh market cap valuation money raised, right?
00:26:05.279 --> 00:26:07.119
You they're at least in that consideration.
00:26:07.279 --> 00:26:12.000
Microsoft obviously one of the biggest and best in the country or sorry, in the world, right?
00:26:12.160 --> 00:26:34.160
When you are starting to see fine tuning as a service, that is not um about model specialization, that is about the uh the impending uh reality of subscription models going away in enterprise companies having to look at that, right?
00:26:34.240 --> 00:26:44.000
If you were paying for, as an example, if you were paying for 10,000 licenses of Microsoft GitHub Copilot, and then all of a sudden they switched over to us at billing, right?
00:26:44.160 --> 00:26:45.759
Are you gonna continue to do that?
00:26:46.000 --> 00:26:46.960
I don't know, right?
00:26:47.039 --> 00:26:49.839
This is where you're gonna have to start looking at these alternatives.
00:26:50.160 --> 00:26:52.720
So here is your Monday move.
00:26:52.960 --> 00:26:55.920
You have to build the spend router, right?
00:26:56.000 --> 00:27:07.759
Whether you look at some of those other services, uh, whether you start doing this and duct taping it yourself internally, or whether you are just preparing for when the big labs offer this, you have to start putting this in place.
00:27:07.920 --> 00:27:10.079
You don't just cut AI blindly.
00:27:10.160 --> 00:27:12.000
That is the absolute worst thing you can do.
00:27:12.160 --> 00:27:15.599
And unfortunately, a lot of companies are doing this.
00:27:15.759 --> 00:27:17.119
So don't do that.
00:27:17.279 --> 00:27:17.759
Uh, right.
00:27:17.839 --> 00:27:24.240
You have to start classifying work by value, risk, and complexity and start using some of these cheaper open source models.
00:27:24.319 --> 00:27:29.519
Again, go through all your due diligence on the privacy data, uh, you know, IP side, all that.
00:27:29.599 --> 00:27:30.079
Uh, right.
00:27:30.160 --> 00:27:45.279
But you should be looking at these tuned specialists, routers, caps, and frontier escalation because the winners ultimately are going to be the ones, I think, like Coinbase that are finding ways to actually spend more tokens and to use more tokens.
00:27:45.359 --> 00:27:56.160
I don't think there was anything wrong per se with the token maxing era, but it was more of doing it irresponsibly on the API side and using the most expensive models, right?
00:27:56.240 --> 00:28:09.680
I do think that you have to have some experimentation on these models that are a fraction of the cost, and you still get maybe, you know, 85 to 95 percent of the performance of the state-of-the-art models.
00:28:10.079 --> 00:28:17.359
So here is your seven-step playbook uh to control runaway AI spend as we wrap today's show.
00:28:17.519 --> 00:28:19.519
Number one, you have to meter first.
00:28:19.680 --> 00:28:24.000
So you have to get token spend visibility by team, task, and model, right?
00:28:24.079 --> 00:28:29.279
So, similarly, how I said, hey, right now I'm going through personally 2.5 billion tokens a week.
00:28:29.440 --> 00:28:30.880
I can break that down by project.
00:28:30.960 --> 00:28:34.000
I actually build a tool inside of Codex that helps me do that, right?
00:28:34.160 --> 00:28:35.680
You can be doing these things as well.
00:28:35.839 --> 00:28:44.000
You have to start seeing where your tokens are going, who is spending them, uh, what task, what model, and what are the outcomes as well.
00:28:44.079 --> 00:28:45.279
Number two, set budgets.
00:28:45.440 --> 00:28:51.119
After you do that, you have to be able to cap spend per team, user, and workflow with overage alerts.
00:28:51.279 --> 00:28:54.319
That doesn't mean you cut use, right?
00:28:54.480 --> 00:28:59.279
So let's say you first meter and you say, here's what we're doing on our subscriptions.
00:28:59.599 --> 00:29:05.279
What is that uh, what is that spend that you're gonna have to do to get the same outcomes, right?
00:29:05.440 --> 00:29:10.880
So don't cut budget without protecting your outcomes, all right.
00:29:10.960 --> 00:29:14.079
Number three, you have to start learning to default to cheap, right?
00:29:14.160 --> 00:29:28.160
I'm not saying the cheapest models, but now we do have models that are extremely capable, open weight models like GLM 5.2, uh, you know, like uh Kimmy um 2.7, right?
00:29:28.319 --> 00:29:30.960
We're gonna, I'm sure, have some other things from Quen.
00:29:31.200 --> 00:29:32.400
Uh, they have some great models.
00:29:32.559 --> 00:29:34.799
Deep Seek, I'm sure we'll come up with another good one.
00:29:34.960 --> 00:29:45.519
Again, go through all your proper uh IP privacy security, but you have to start looking at these maybe open source, open weight models, whether you're running them locally or paying for them on the API side.
00:29:45.599 --> 00:29:50.000
They're normally a fraction of the price as the closed uh frontier models.
00:29:50.160 --> 00:29:52.319
Number four, you have to route by difficulty.
00:29:52.480 --> 00:30:00.319
You shouldn't be, as an example, having Fable five on extra ultra reasoning, whatever it's called inside Cloud Code.
00:30:00.400 --> 00:30:03.680
You shouldn't be doing that to help you write better emails, right?
00:30:03.839 --> 00:30:08.079
You have to send the right type of tasks to the right models.
00:30:08.240 --> 00:30:12.480
Uh, then you need to understand your get proper uh caching.
00:30:12.640 --> 00:30:18.559
You need to trim and reuse repeated prompts and start lean sessions per task, right?
00:30:18.640 --> 00:30:24.079
Even simple things like if you're working on the front end using things like projects, right?
00:30:24.160 --> 00:30:29.759
Uh certain uh, you know, certain providers do things better than others, right?
00:30:29.839 --> 00:30:34.319
As an example, Codex is great at compaction, auto compaction, it's really good.
00:30:34.480 --> 00:30:45.680
Um, you know, Claude recommends as chats get longer to start new chats, uh, to save on cost, but you have to start doing those kind of non-technical best practices as well as caching and trimming.
00:30:45.839 --> 00:30:48.000
Number six, fine-tune the repeats.
00:30:48.160 --> 00:30:57.039
Yes, if especially I think if you are a uh Microsoft organization, I do assume that we're gonna start seeing fine-tuning as a service, right?
00:30:57.200 --> 00:31:07.839
I think Thinking Machine Labs and Microsoft uh offering this leads me to believe we're gonna see this probably within uh nine months from all of the big players.
00:31:08.000 --> 00:31:13.200
You have to start looking at and understanding what it's what are we gonna fine tune, right?
00:31:13.279 --> 00:31:19.200
What type of model is going to do 80% of our AI use for certain teams?
00:31:19.359 --> 00:31:27.680
And well, if you fine-tune a model, it might be a little bit of work and a little bit of cost up front, but you have to be able to math the math because chances are it's gonna save you a lot in the long run.
00:31:27.839 --> 00:31:31.440
And then last but not least, escalate on purpose.
00:31:31.759 --> 00:31:40.319
Call the frontier model only to close hard problems, especially if that frontier model is not included in a subscription.
00:31:40.400 --> 00:31:43.599
And I say that as a very timely reminder.
00:31:43.759 --> 00:31:48.160
Yes, Fables, uh Fable Five, a couple more hours, and it's gone.
00:31:48.319 --> 00:31:57.039
We'll see what happens with open AI, uh, GPT-5.6, Soul, uh, and Soul Ultra, they did say it's gonna be included uh inside codex.
00:31:57.279 --> 00:31:59.200
So who knows what's happening?
00:31:59.279 --> 00:32:06.880
But all I know right now, as we wrap up today's show, the future of AI is token efficiency.
00:32:07.039 --> 00:32:19.839
And it is no longer uh, if you are the decision maker, six months ago, regardless of what tool, what system, the right move was to use as much AI as possible.
00:32:20.000 --> 00:32:34.960
And that still may be the case, especially if you're an open AI organization, because they are pretty much the only player right now that hasn't drastically uh either cut uh the subscription limits or that hasn't gone to true metered access.
00:32:35.119 --> 00:32:43.920
But chances are some of your team, some of your organizations most needed, most used AI tools have converted to metered pricing.
00:32:44.160 --> 00:32:46.160
So that doesn't mean stop.
00:32:46.720 --> 00:32:48.960
That just means be smarter.
00:32:49.119 --> 00:32:54.400
Don't stop experimenting and well, follow us daily because we're gonna be helping you every step along the way.
00:32:54.559 --> 00:32:56.240
I hope this one was helpful.
00:32:56.400 --> 00:33:09.119
If so, please make sure to go to starthirseries.com, go sign up for our community, uh, go network with other people who are leaders in the AI industry, uh, figuring it all out there.
00:33:09.200 --> 00:33:16.240
Uh also, if this was helpful, please subscribe to the show on Spotify and on Apple Podcasts.
00:33:16.319 --> 00:33:17.119
Thank you for tuning in.
00:33:17.279 --> 00:33:20.480
Hope to see you back tomorrow and every day for more everyday AI.
00:33:20.559 --> 00:33:21.519
Thanks, y'all.
