WEBVTT

00:00:00.060 --> 00:00:08.660
Speaker 0: 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.

00:00:09.000 --> 00:00:11.360
Speaker 0: Rightfully so because it makes no sense.

00:00:12.140 --> 00:00:24.560
Speaker 0: Tokenmaxing in case you don't know is simply about using or burning as many AI tokens per year whatever time period.

00:00:26.380 --> 00:00:31.540
Speaker 0: The idea from a company perspective, because this is term coming form the enterprise world.

00:00:32.080 --> 00:00:44.240
Speaker 0: 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.

00:00:44.360 --> 00:00:48.380
Speaker 0: Just like Codex at these other tools you can get work done for them.

00:00:48.560 --> 00:00:50.520
Speaker 0: I got courses on Cloudcode in Codex.

00:00:50.600 --> 00:00:53.140
Speaker 0: if want to learn more they're really indebted.

00:00:53.220 --> 00:00:54.440
Speaker 0: show some tips and tricks.

00:00:54.540 --> 00:01:02.940
Speaker 0: but the idea it's use this tools to burn or uses many tokens possible because will give great outputs right?

00:01:03.760 --> 00:01:03.840
Speaker 0: No!

00:01:04.580 --> 00:01:07.800
Speaker 0: As mentioned... are valuable.

00:01:08.060 --> 00:01:33.260
Speaker 0: 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.

00:01:36.440 --> 00:01:56.460
Speaker 0: 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.

00:01:57.000 --> 00:02:05.300
Speaker 0: 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.

00:02:05.680 --> 00:02:08.539
Speaker 0: You want to understand and review that code?

00:02:09.039 --> 00:02:12.720
Speaker 0: It's not about spitting out as much code as possible.

00:02:12.780 --> 00:02:15.000
Speaker 0: It never was even before AI.

00:02:15.340 --> 00:02:23.920
Speaker 0: 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.

00:02:24.020 --> 00:02:25.060
Speaker 0: And it's not different.

00:02:25.140 --> 00:02:25.920
Speaker 0: with AI.

00:02:26.700 --> 00:02:28.020
Speaker 0: The quality matters.

00:02:28.140 --> 00:02:35.760
Speaker 0: And I know that this seems to be something not all companies would agree with these days, but yes it does.

00:02:36.160 --> 00:02:47.980
Speaker 0: 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?

00:02:48.040 --> 00:03:01.380
Speaker 0: 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.

00:03:02.940 --> 00:03:11.080
Speaker 0: So of course we need that human control to use AI efficiently and get good results!

00:03:11.900 --> 00:03:37.080
Speaker 0: 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.

00:03:37.500 --> 00:03:46.620
Speaker 0: 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.

00:03:47.720 --> 00:03:50.440
Speaker 0: Now turns out this isn't working too well.

00:03:51.100 --> 00:04:09.740
Speaker 0: 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.

00:04:09.820 --> 00:04:17.880
Speaker 0: so they had a budget of tokens They wanted to pay for or use And they burn threw it within a few months now.

00:04:18.019 --> 00:04:20.399
Speaker 0: Of course And that's important, I think.

00:04:20.880 --> 00:04:31.320
Speaker 0: One reason is the budget was likely set in a year and one would assume or at the end of early twenty-six.

00:04:31.420 --> 00:04:34.540
Speaker 0: Then we had an agentic coding takeoff.

00:04:37.040 --> 00:04:38.320
Speaker 0: So that happened.

00:04:38.940 --> 00:04:54.860
Speaker 0: 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.

00:04:55.220 --> 00:04:59.220
Speaker 0: and then those tools cloud code Codex on which I have Those courses i mentioned Which are great.

00:04:59.600 --> 00:05:04.680
Speaker 0: The Tools in the Courses Those also got better and used those models really efficiently.

00:05:04.760 --> 00:05:14.260
Speaker 0: 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.

00:05:14.420 --> 00:05:25.940
Speaker 0: 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.

00:05:26.480 --> 00:05:39.040
Speaker 0: 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.

00:05:39.100 --> 00:05:40.920
Speaker 0: and of course the longer a session runs.

00:05:40.960 --> 00:05:42.000
Speaker 0: The more tokens at burns.

00:05:42.420 --> 00:05:54.940
Speaker 0: 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.

00:05:55.980 --> 00:05:59.260
Speaker 0: But nonetheless, Uber burned through its entire budget.

00:05:59.360 --> 00:06:05.120
Speaker 0: Now if they were getting amazing results They would surely increase their budget.

00:06:05.600 --> 00:06:08.020
Speaker 0: but doesn't look like that is what happened.

00:06:08.600 --> 00:06:14.840
Speaker 0: and Nvidia executive also said the cost of compute Is far beyond the costs of employees.

00:06:14.960 --> 00:06:20.420
Speaker 0: So it's more expensive right now to use AI than to use humans?

00:06:20.640 --> 00:06:25.120
Speaker 0: Of course you could say Doesn't matter if AI is ten times as productive.

00:06:25.200 --> 00:06:30.840
Speaker 0: As a human employee, it's fine If It's Ten Times or eight times as expensive right?

00:06:31.180 --> 00:06:36.840
Speaker 0: Maybe it would even be fine if it were fifteen times as Expensive because it can get Even better.

00:06:36.980 --> 00:06:44.700
Speaker 0: whereas for the Human Productivity...it Can also increase but probably not as sharply as that of AI.

00:06:45.920 --> 00:07:20.000
Speaker 0: 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.

00:07:20.060 --> 00:07:22.320
Speaker 0: what Will Likely Come Next In A codebase?

00:07:22.700 --> 00:07:25.040
Speaker 0: Which Future Capabilities May Be Needed.

00:07:26.180 --> 00:07:28.060
Speaker 0: AI models are missing, of course.

00:07:28.460 --> 00:07:36.400
Speaker 0: So it's so stupid from many different angles to compare productivity with human productivity.

00:07:36.640 --> 00:07:40.000
Speaker 0: and the first companies are seeing that I think.

00:07:40.080 --> 00:07:41.980
Speaker 0: which is why all this token maxing?

00:07:42.020 --> 00:08:02.860
Speaker 0: 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.

00:08:03.460 --> 00:08:05.480
Speaker 0: things will settle down a bit more.

00:08:05.860 --> 00:08:09.620
Speaker 0: AI is here to stay and AI is useful, it's a useful tool!

00:08:09.940 --> 00:08:29.980
Speaker 0: 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.

00:08:33.159 --> 00:08:35.659
Speaker 0: And it can even be useful for wipe coding.

00:08:35.720 --> 00:08:38.679
Speaker 0: if you need a little tool that just does something.

00:08:38.980 --> 00:08:51.720
Speaker 0: 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!

00:08:52.500 --> 00:08:57.420
Speaker 0: There are many great use cases for AI and it's a technology that is here, will stay in.

00:08:57.520 --> 00:08:59.160
Speaker 0: It'll become better of course!

00:08:59.400 --> 00:09:03.120
Speaker 0: And nobody knows what the case would be in ten years or so.

00:09:03.240 --> 00:09:15.900
Speaker 0: 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.

00:09:16.160 --> 00:09:43.740
Speaker 0: 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?

00:09:47.220 --> 00:10:37.340
Speaker 0: 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?

00:10:37.460 --> 00:10:41.040
Speaker 0: I'm sure it'll actually lead to more work.

00:10:41.100 --> 00:10:43.800
Speaker 0: that has been a case with those technological breakthroughs.

00:10:44.640 --> 00:10:47.940
Speaker 0: And like them we just don't see how future roles look.

00:10:48.200 --> 00:10:51.940
Speaker 0: but when you take a look at coding... We're not even close!

00:10:52.940 --> 00:11:00.280
Speaker 0: where you would want to let AI write all the code and not care about it at all for any serious product.

00:11:00.780 --> 00:11:07.000
Speaker 0: At least I definitely wouldn't, and i think any company that would make grave mistakes.

00:11:07.620 --> 00:11:19.180
Speaker 0: 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.