Companies are starting to realize that there's maybe too much AI hype...

00:00:00: Let's talk about one of the most stupid trends we saw over a couple weeks and months, which as it seems is coming to an end already.

00:00:09: Rightfully so because it makes no sense.

00:00:12: Tokenmaxing in case you don't know is simply about using or burning as many AI tokens per year whatever time period.

00:00:26: The idea from a company perspective, because this is term coming form the enterprise world.

00:00:32: The idea really that you want to incentivize your employees For example, through tools like cloud code and just as a side note that is useful tool.

00:00:44: Just like Codex at these other tools you can get work done for them.

00:00:48: I got courses on Cloudcode in Codex.

00:00:50: if want to learn more they're really indebted.

00:00:53: show some tips and tricks.

00:00:54: but the idea it's use this tools to burn or uses many tokens possible because will give great outputs right?

00:01:03: No!

00:01:04: As mentioned... are valuable.

00:01:08: As a developer I believe you need to be able work with these tools but use them as assistance, the idea behind token maxing or incentive is that prompt after prompt that you look at the output as little or not, of course.

00:01:36: We've heard about companies having internal leaderboards where people who spend most tokens would be on top and potentially get some rewards... And of course I'm mostly talking about AI being used for development here, because that's where i am coming from.

00:01:57: But I would say it doesn't make sense in any context, but especially if we're talking about using AI for writing code or for generating code.

00:02:05: You want to understand and review that code?

00:02:09: It's not about spitting out as much code as possible.

00:02:12: It never was even before AI.

00:02:15: It wasn't a good idea to measure the productivity of a developer by the lines of code they can write on a given day.

00:02:24: And it's not different.

00:02:25: with AI.

00:02:26: The quality matters.

00:02:28: And I know that this seems to be something not all companies would agree with these days, but yes it does.

00:02:36: if you go down the in-the end wipe coding rabbit hole and You have AI Generate all that spaghetti code and You totally lose track of what's going on?

00:02:48: And you don't understand What is about then your not able to dive into the code manually at some point because just too much then you lost, because AI is far from perfect.

00:03:02: So of course we need that human control to use AI efficiently and get good results!

00:03:11: That's why I've been saying for all my videos but it's a tool not the replacement of developers no matter how much certain companies are wishing And of course, the entire idea behind token maxing in the end is that hopefully from the company's perspective you can get to a point where your employees are using AI so much.

00:03:37: they're producing lots amazing output and as a company say great now we don't need all those employees anymore or at least less.

00:03:47: Now turns out this isn't working too well.

00:03:51: There is this report about Uber, which got quite popular on X for example over the last weeks where The uber COO and I think also their CTO in the end mentioned that they burned through Their entire twenty-six AI budget In four months.

00:04:09: so they had a budget of tokens They wanted to pay for or use And they burn threw it within a few months now.

00:04:18: Of course And that's important, I think.

00:04:20: One reason is the budget was likely set in a year and one would assume or at the end of early twenty-six.

00:04:31: Then we had an agentic coding takeoff.

00:04:37: So that happened.

00:04:38: And, uh...that happened of course because certain models like Opus four point five but also GPT Five point Four or Codex before that got really good Or it got better especially at following instructions At the end of last year.

00:04:55: and then those tools cloud code Codex on which I have Those courses i mentioned Which are great.

00:04:59: The Tools in the Courses Those also got better and used those models really efficiently.

00:05:04: And of course, all the other tools like Pi which is an amazing coding agent ends on now The combination that led to more usage off these tools.

00:05:14: But since we're talking about a genetic coding here where these tools or wear the models in this tool think and use tools, call tools do searches analyze the search results.

00:05:26: that all costs way more tokens than We or these companies used AI last year where it was shorter sessions, not so many long running agentic sessions.

00:05:39: and of course the longer a session runs.

00:05:40: The more tokens at burns.

00:05:42: So the calculation that happened in twenty-twenty five has nothing to do with the reality Of how ai is being used in twenty twenty six With those enhanced agendic coding models And the tools around them.

00:05:55: But nonetheless, Uber burned through its entire budget.

00:05:59: Now if they were getting amazing results They would surely increase their budget.

00:06:05: but doesn't look like that is what happened.

00:06:08: and Nvidia executive also said the cost of compute Is far beyond the costs of employees.

00:06:14: So it's more expensive right now to use AI than to use humans?

00:06:20: Of course you could say Doesn't matter if AI is ten times as productive.

00:06:25: As a human employee, it's fine If It's Ten Times or eight times as expensive right?

00:06:31: Maybe it would even be fine if it were fifteen times as Expensive because it can get Even better.

00:06:36: whereas for the Human Productivity...it Can also increase but probably not as sharply as that of AI.

00:06:45: But we're all so not near to these numbers ten acts fifteen acts because again the number of lines of code generated is not a good measure and we need human employees with their experience, Connection to other departments in a company with all those nuances that make up the job of course, With All The Trust That Is Assigned To A Human and Of Course Also With Their Deep Understanding Of What Makes A Good Code Base.

00:07:20: what Will Likely Come Next In A codebase?

00:07:22: Which Future Capabilities May Be Needed.

00:07:26: AI models are missing, of course.

00:07:28: So it's so stupid from many different angles to compare productivity with human productivity.

00:07:36: and the first companies are seeing that I think.

00:07:40: which is why all this token maxing?

00:07:42: here You can read about more and more companies like Amazon, Meta and many others that are cutting back on their token leaderboards or on the token maxing approach here.

00:08:03: things will settle down a bit more.

00:08:05: AI is here to stay and AI is useful, it's a useful tool!

00:08:09: It can make you more productive, its great for doing extra researches, great for producing that boilerplate code or also the non-boiler platecode but based on clearly defined specifications with human review ideally based some code base which was at least shaped by humans.

00:08:33: And it can even be useful for wipe coding.

00:08:35: if you need a little tool that just does something.

00:08:38: You need to get done right now, which don't plan on publishing the world where you do not care about all of your bugs and will not add much features... Which is great too!

00:08:52: There are many great use cases for AI and it's a technology that is here, will stay in.

00:08:57: It'll become better of course!

00:08:59: And nobody knows what the case would be in ten years or so.

00:09:03: but right now I really hope things will settle down more and we can use AI to do something useful too But not that magic thing which changes everything.

00:09:16: We get rid off all jobs replace employees within the next twelve months And it looks like, probably for publicity reasons though that even our favorite tech CEOs Sam Altman and especially also Dario M. O'Day are kind of retreating regarding those pretty strong statements on how soon AI will replace all white-collar work right?

00:09:47: pretty wrong about AI's economic impact and an anthropic CEO, Dario M. Odei who not so long ago mentioned that most or all of white collar work will be replaced by AI relatively soon now says automation may actually expand the work people do Probably because their PR department told them that whilst it's amazing for selling tools to companies when they say how many employees can replace, its not so amazing if the entire world turns against near in the future, will AI replace all white collar work?

00:10:37: I'm sure it'll actually lead to more work.

00:10:41: that has been a case with those technological breakthroughs.

00:10:44: And like them we just don't see how future roles look.

00:10:48: but when you take a look at coding... We're not even close!

00:10:52: where you would want to let AI write all the code and not care about it at all for any serious product.

00:11:00: At least I definitely wouldn't, and i think any company that would make grave mistakes.

00:11:07: but as it seems companies also are hopefully starting to realize that AI is better used as a great tool instead of an all-in do everything solution.