Microsoft Admits it was Wrong About AI
Video Overview & Insights
For years, we were told AI would replace programmers, office workers, and eventually most white-collar jobs. But behind closed doors, the companies leading the AI revolution are running into a problem they never expected. AI is becoming too expensive to replace humans.
My ex CTO told me. Ohh a QA guy is winning. I told him. Dude. Do you want to spend more money? 😂. Im saving the company money on tokens.
Microsoft has started pulling back internal AI tools, Uber exhausted its AI budget in just a few months, and even the biggest tech companies are struggling to justify the costs.
This video explores why Microsoft's AI strategy suddenly changed, how Uber's AI rollout spiraled into a budget disaster, and why "AI agents" are costing companies far more than they anticipated.
Who would have thought that the creators of the agents will make the agents work in a way that is beneficial to the maker? Gasp
We examine token costs, inference expenses, Nvidia's role in the AI boom, GitHub Copilot, Claude Code, and why the economics of artificial intelligence may be very different from what CEOs promised. We also look at the hidden costs of AI automation, why replacing human workers isn't as simple as it sounds, and how the industry's biggest investment could be running into a fundamental economic wall.
Has the AI revolution hit its first major reality check, or is this just the beginning of a much larger shift in how businesses use artificial intelligence?
At about 5'30" Josh mentions 1 prompt = 1 token. That's not even close. Even a simple prompt to a LLM can burn through hundreds of tokens based on the words in, the response required and even the number of words each way. In fact, these things are so wordy (can't imagine why...) that it's hard to keep them to one - five word answers e even when a SINGLE word (yes/no) is required.
EDIT: correction, perhaps Josh is referring to a prompt USING an agent - but agents are little more than an LLM on top of a tuned model.
00:33 - The Reveal (Firing the Digital Worker)
03:51 - The Tokenmaxxing Trap
This is good news a couple of hundred trillion watts of electricity too late.
05:17 - The Agentic Scam
06:53 - The Inference Wall
There is a simple solution to this - as soon as AI causes the cost of living to drop significantly, it will start to look much clearer. It is simply a matter of lowering the salaries of high-tech employees to a wage that justifies the work in relation to an AI Agent. In such a situation, I do not foresee companies cutting back on manpower to ensure code quality. Quite the opposite... because if a lot of code is written, naturally, the demand for more "human in the loop" will only increase significantly. Again, this is the logical and likely scenario in a situation where the cost of a human employee will be much lower than it is today - which, as I said, is a reasonable thing in a world where the cost of living is expected to drop because AI will make everything cheaper.
09:05 - The Margin Collapse
10:54 - The Efficiency Paradox
AI greed did NOT work.
Who would believe that... 😊
12:07 - The Reckoning (The Human Victory)
Narrated by: Josh Risser
My company too is pushing for use of Ai. Result, I have set Opus as my default model and throw every random query towards it......
🔔 Don't forget to SUBSCRIBE! 🔔
SUGGEST A TOPIC: https://bit.ly/suggest-an-infographics-video
AI Slop.
💬 Come chat with me: https://discord.gg/theinfoshow
🔖 MY SOCIAL PAGES
We went from 'AI is going to replace everyone' to 'Who left the AI running and blew our entire annual budget?' in record time. It turns out autonomous agents getting stuck in expensive troubleshooting loops is the tech equivalent of a plumber charging you by the hour just to drop his wrench. Humans are officially the budget option!
TikTok ► https://www.tiktok.com/@theinfographicsshow
Facebook ► https://www.facebook.com/TheInfographicsShow
AI didn’t fail because it was dangerous — it failed because it’s too expensive to run.
📝 SOURCES: https://pastebin.com/tgKL2U1b
All videos are based on publicly available information unless otherwise noted.
Sounds like infinite money glitch doesn't even manifest with complex AI. Stop trying to escape entropy
More User Perspectives
yo what up doodlebob
@aboohoowhen ya train on trash, ya get trash...
@NoahJoeKingNice information ❤❤Cute Human employees making comeback 😘🥰❤
@rajadon2071Just like internet.
When you paid for each Mb. Internet was very basic, almost text only medium.
Moment you started seeing 100mb/s, unlimited data internet... it exploded into what we use now.
AI will have to be almost free for normal users like internet. 100$ a month and 95% of people wont even make a dent, but 5% are companies that truly use AI.
Unless we hit this zone, as long people look at red line of daily tokens ?
It wont explode in popularity.
AI is here to stay. Use it for the correct cases and derive real value. Misuse it and everyone suffers. Humans are consumers who pay.
AI is not the consumer, it's a Cost incurred.
Be wise and you will secure real Profits . ❤
They're not making A.I. to make more money, they're doing it to reach full controll of every part of society and industries they'll put there a.i. in.
@hervelescalmel620MACROSLOP
@Gradius2Just turn off the testing/debugging part and you save so much
@waffles87"The junior’s output doubles when the pay stays around the same" this is the problem, AI in itself is not as "slaveable" as humans in the eyes of capitalism
@bvelay4431024Lol maybe because Microsoft is shit on AI.
@coin777This is so satisfying man ☺️☺️☺️☺️
@vicsac2946Tokenprice will not drop. Anthropic just raised the price. Same text now "cost" 30% more, because they change the tokenizer
@antarcticSnowflakeMoye moye
@abhishekgourav6144I see people showing me code at work they are wanting to use generated by AI and its like wtf that is horrible and wrong don't do that.
@SwanzoSo, I have some task need to be done and previously, I check how to and actually doing it by copy past and edit.
Now, even I know how to find the way to do it, I just ask AI to do that. And it burns token, and the company still pays me. Luckly, I am in a low cost region so I think the company still can afford it.
I guess in a short years, when the computational power goes up, the AI cost will go down, but at what cost. All the hardware just becomes worthless
Why you are always after microsoft 😂
@akshayshetye8718I work for a Singaporean firm as a senior 3d visualiser. I work offshore. Their goverment is pushing HARD on ai adoption offering incentives, credits etc. My boss meanwhile had been trying to push ai adoption for a very long while now. Setting up an ai task force (consists of all designers btw), and basically rushing and pushing us to adopt ai into paperwork, drawings, design and of course 3d. I resigned from the ai task force because I dont want to do extra work tbh. Safe to say, my boss is very unhappy but I still get to keep my job because my skills still has value. He keeps seeing those glorified demos and telling us how ai can speed things up which is all bs IMHO cuz reality is very much different. Ps : I do use ai to assist in generating textures and models so I'm not anti Ai. I use it when its useful. Otherwise its just a parasite of my time and energy.
@kywenmook284Humans...so intelligent and yet so dumb. 😆
@eideardpeschak7546I have been working with Ai since 2009. It ain't all that.
@PilotVBallWhy are companies not rewarding employees for using the least amount of tokens for the most amount of relevant work accomplished? This reminds me in the 70s and 80s when programmers were paid by the line of code written, instead of the opposite.
@MP-mn6weIf I was tokenmaxxing, I would make the AI write AI to do the most mundane code, then ask AI where I should use AI in the code
@JoeStuffzAltim not convince they did not know this before
@SergeC0Are anti AI videos all you do now?
@karlsilveroneIt is funny to think that people assume the cost of ai is low but when you look at what it actually cost to produce results its really not that cheap. Not only does it require expensive equipment that need to be upgraded ever few years on top of the basic resources like electricity and water is not free and these ai companies need to actually make money at some point. The only way to do so is to charge. Just imagine your own water bill then multiply it multiple time and that what one person using ai cost. until they figure out a way to effectively cool these data centers ai will never truly be cheaper. They are at least trying to figure out the electricity problem but if they keep using fresh water like they are that resource will just become more expensive. Not to mention what that will mean for everyone but the people pushing this dont care and know they can afford water no matter how much it cost.
@JoeyGGbroFirst of all its not about budget option, llms cannot hold context of the whole project
@zodder13I dont understand the token consumption process
@Varifyrbro have to show him recording directly in front of the mic to prove that the audio isnt AI 😂
@rax_low_spenderNow coders must increase their 20 times
@ishankumar2392Come on everyone back to work, It won't changed our lives.
@Therealgame1I like how you had to start showing AI videos of a person talking at the start of your videos just so people would stop asking if this is an AI voice channel.
@lolpaladinsAnd yet they push so hard for the data centers. And trying to pass it off as "it's for AI"🙄
@briankennedy5578And the AI bubble is starting to pop
@Refusenik6969Great so our future we have to look forward to is managing 100 slightly retarded/hallucinating AI bots each ... postal time!
@mikegwillisSource?
@BeiCorn2316The caveat is, or at least should be, that AI reduces worktime and fatigue making the process itself faster, easyer and more efficient. There is no real way to measure, if the workers achived more in less time, especially if heavy token use had such positive bias. But in theory this would allow workers to be faster. The problem is that the current workload structure is, in most cases, not set in stone. The only work that needs to be done is the one that has a deadline or is already due. The whole system is fluid, so if you do 2 years worth of work in six months, you may get a pat on the back, but if you were to also use the resources needed for that amount of work, they might even blame you for being wastefull.
@akso2166Sorry? Yeah yeah. As long as we don’t pay for your mistake……….oh wait!
@wakaneutthis is propaganda
@chadgrant8870How ironic... the driving factor behind the forced integration of AI, greed, is the thing that ultimately shuts the whole thing down. "...for it is easier for a camel to pass through the eye of a needle than for a rich man to path through the gates of heaven" ...who knew? Oh wait... LITERALLY EVERYBODY... for 2000 odd years! I love science!
@passion_proh-jectsI know this sounds like a stupid question, but how is Uber spending billions a year writing code to begin with? Isn’t the app already built and running? What is this code “improving”?
@lukethompson5558