Oscar Cardoso

Poor prodigy kid (UnEncouraged, UnRecognized UnTapped or Channeled Correctly) WorldObserver FreeThinker {WantMinds NotFollowers} NoBots

[Oscar Cardoso @vryand]

One last shot at this: @jack
Clear Signals App (MVP Done)
https://t.co/8K9AtuRJyh
https://t.co/WAGPjl0PUF
https://t.co/FMwvfURFgz
https://t.co/qyv4WnVpPB

[Oscar Cardoso @vryand]

@phipps By the Human-Agent-LLM issue I mean that I see many are 'Token-maxin' or 'Generating xK of lines code' per day and I still do not see mush said or done about the quality or of safeguards, agents are still having limited context windows and the LLMs are not utilized efficiently

[Oscar Cardoso @vryand]

This is good as a Agentic automation audit (needed now) to lowering the cost of many over engineered processes many got accustomed to create when the price of AI was low.

Only thing is that is does not correct the initial Human-Agent-LLM issue from the start of new projects.

[Travis Phipps @phipps]

https://t.co/Ddv8M2Slvp

[Oscar Cardoso @vryand]

AI solutions = Old man with Alzheimer's + Dementia on drugs that forgets where, who, what it was doing put in charge of sensitive private information replacing...
Now more expensive then the young workforce...
No actual QA testing (test it self)

What is the worse it can happen?

[Oscar Cardoso @vryand]

500M a 150M ahorar 350M... que no daria por solo 1 a cambio de ahorarles esos 350M-1M

[Guillermo Izquierdo @gizquierdo_dev]

Un consultor citado en Axios: un cliente gastó $500M en un mes solo en Claude porque nadie puso límites de uso por empleado. Es UN cliente. Eso multiplicado por 12 son $6B de run-rate extra para Anthropic.

[Guillermo Izquierdo @gizquierdo_dev]

Un consultor citado en Axios: un cliente gastó $500M en un mes solo en Claude porque nadie puso límites de uso por empleado. Es UN cliente. Eso multiplicado por 12 son $6B de run-rate extra para Anthropic.

[Oscar Cardoso @vryand]

@theonechrisyep @leonpalafox Comparto este sentimiento sobre la diferencia economica

[Christian Yepez @theonechrisyep]

@leonpalafox Veo la brecha tambien en cantidades a gastar en tokens. Leo aa devs de EUA gastando $4K USD al mes en pruebas y creando cosas. Acá solo hacen lo que permiten los límites gratuitos de alguna LLM o el tier mas bajo de Claude.

[Oscar Cardoso @vryand]

@phipps This is good as a Agentic automation audit (needed now) to lowering the cost of many over engineered processes many got accustomed to create when the price of AI was low.

Only thing is that is does not correct the initial Human-Agent-LLM issue from the start of new projects.

[CXOTALK @cxotalk]

@SecretCFO If your company is using LLMs for AI agents, you're probably seeing significant token cost and its growth. You're probably seeing sky rocketing token cost if you're attempting or doing agentic workflows.
#AgenticAI #AIagents #AgenticWorkflows

[Oscar Cardoso @vryand]

@RoshanMayengba I just created a technique to cut the token usage down to 30% (70% savings)

DM me if you find people that this can help them.

[Vaibhav Sisinty @VaibhavSisinty]

This guy on Reddit burned 1.15 BILLION Claude tokens in a single month

And what he learned will save you thousands of dollars on AI. 🤯

5 takeaways worth saving:

→ Prompt caching got quietly nerfed: Anthropic cut cache time from 60 minutes to 5, silently raising production costs by 30-60% for most users.

→ Output tokens cost 5x more than input: Stop asking AI for full text, ask for IDs or numbers and map them in your code — he cut his output bill by 60% doing this.

→ JSON is a silent token killer: Every bracket, quote, and comma eats tokens, making the same data cost 2x more in JSON than in plain text or markdown tables.

→ Opus 4.7's tokenizer secretly raised your bill: The new tokenizer generates up to 35% more tokens than Opus 4.6 for the exact same input, and nobody is talking about it.

→ You're using the wrong model for most tasks: Haiku is 5x cheaper than Opus and good enough for 80% of real work, so stop defaulting to the flagship for everything.

The wild part? He runs an AI agent company. This is what he learned by burning real money.

While everyone races to use AI, the smart ones are learning how to use it cheaper.

[Oscar Cardoso @vryand]

@StartupHakk I need to reach out to who wasted 40K.

I can cut that budget down to 12K

[StartupHakk @StartupHakk]

AI shouldn't be a rental. It should be infrastructure you own, answering only to you, as token costs skyrocket. Running 2100 agents cost $40k—budgets are breaking. Flat-rate AI coding subscriptions are ending. Get ready for AI token costs to exceed employee salaries.

[Oscar Cardoso @vryand]

@DavidLinthicum If I help you making those 100K down to 30K, how mush f the 70K savings are you willing to share?

Plus (Extra) it includes one shot prompts that work.

[DavidLinthicum @DavidLinthicum]

Agentic AI systems are facing massive cost overruns as LLM providers hike token prices. What starts as $1k/month can balloon to $100k. Dependence on these LLMs means unpredictable, escalating costs. Is the business value worth it? #AICosts #LLM

[Hedgie @HedgieMarkets]

🦔Tech companies that pushed employees to maximize AI usage are now realizing the math does not work. Microsoft, Meta, and Amazon all set internal targets that pressured workers to use AI tokens aggressively to hit productivity scores. The problem is agentic AI burns up to 1,000 times more tokens per task than a standard LLM query because it loops through multiple steps and self-checks.
OpenClaw's creator Peter Steinberger said his team spent $1.3 million on OpenAI tokens in a single month. Nvidia CEO Jensen Huang told his engineers they should be consuming AI tokens worth at least half their annual salary every year. The behavior has its own name now, "tokenmaxxing."

My Take
The cost trajectory works backwards from how the labs sold it. Per-token prices have fallen, but the number of tokens each task consumes has climbed faster, and the all-in spend keeps going up release after release. Agentic AI is the worst offender because the model talks to itself, second-guesses itself, and runs the same logic three times before landing on an answer. Goodhart's Law also shows up clearly here. When AI usage became the performance review metric, employees started using AI to inflate the metric, not because the task needed AI.

OpenAI and Anthropic are losing roughly $2 for every $1 of revenue, and the only way the math fixes itself is by raising prices or capping consumption per enterprise contract. Both moves slow the revenue growth the labs need to show on the IPO roadshow. Goldman Sachs and the underwriters know this, which is why SpaceX's S-1 came out before OpenAI's. Whichever AI lab files first gets the cleaner narrative, and whoever files second has to explain why their largest enterprise customers just started rolling back token consumption. The companies pushing tokenmaxxing internally are now the same companies signaling cost pressure externally, and that contradiction is going to show up in earnings the moment these labs start reporting publicly.

Hedgie🤗

[Oscar Cardoso @vryand]

The new workforce

[Oscar Cardoso @vryand]

The new gold/gas/oil of the new age economy

[Oscar Cardoso @vryand]

Share with me your experience with the Development IDE agent when it comes to the rules, what methods have you tried already.

[Oscar Cardoso @vryand]

I have already tried 5 (single *.md file, group of *.md files, cursor folders...) some worked but only for a short time, since my project needed to have more rules every day, then its starts forgetting the rules once again (they are not consistent or 100% reliable)