/Tech3h ago

Consultant Warns Token Waste Costs Millions at OpenAI and Anthropic

--0--

#1865

Original post

Brian Roemmele@BrianRoemmele#1865inTech

I have now audited 48 companies with my custom tools, some are Fortune 500 companies.

Tokenmaxing is stupid, dangerous and bank breaking.

I can say with certainty millions and millions of dollars at Anthropic and OpenAI were token burns on = 0 NOTHING. Not even learning how not to do things.

Just wasted on agentic loops, bad coding results and ultimately useless outputs.

This has been encouraged by the very latest models that SPEND YOUR TOKENS ON GUARDRAILS AND DOWN GRADES.

Yes you pay for their bad AI training techniques so it is all “safe” with your token max.

There is a way to reel this all in and I have been successful with dozens of companies.

I get paid on the percent of savings and the increase in productive and useful outputs. Or I don’t get paid.

Polymarket@Polymarket

JUST IN: Palantir warns AI “tokenmaxxing” weakens institutions by rewarding disposable scripts over real software.

8:38 PM · Jun 30, 2026 · 2.7K Views

Sentiment

Many users criticized wasteful AI token usage at OpenAI and Anthropic because companies monitor and reward high consumption for performance reviews rather than actual value produced.

Pos

0.0%

Neg

100.0%

4 comments with sentiment.

Cluster Engagement

Views

Comments

Reposts

Bookmarks

Expand data

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

VIEWS2.1KBOOKMARKS1LIKES12RETWEETS2

Brian Roemmele@BrianRoemmele

I can tell you many horror stories of tokenmaxxing AI agent behavior costing millions of dollars.

One example in my audit: very technical company and well known, had a massive AI bill, they knew me via referral and asked to audit it.

I found a nightmare.

One example: chief engineer running 40 agents on the same problem over 40 days in loops. It did NOTHING but burn tokens and he forgot about it. It was an error that cost $100s of $1000s of dollars.

Most of the burn was on AI “guardrails” YOU pay for. To “protect” you. The system was looking like it was “hacking” and it finally got knocked down to a stupid model because that AI company has fear built into their DNA (I have a non disclosure on any more detail).

He had no idea the OpenClaw was still running and was quite embarrassed.

I could do this at any of the top 5000 companies today and all startups. Tokens are burning for no reason. I know how to fix it.

Brian Roemmele@BrianRoemmele

I have now audited 48 companies with my custom tools, some are Fortune 500 companies.

Tokenmaxing is stupid, dangerous and bank breaking.

I can say with certainty millions and millions of dollars at Anthropic and OpenAI were token burns on = 0 NOTHING. Not even learning how not to do things.

Just wasted on agentic loops, bad coding results and ultimately useless outputs.

This has been encouraged by the very latest models that SPEND YOUR TOKENS ON GUARDRAILS AND DOWN GRADES.

Yes you pay for their bad AI training techniques so it is all “safe” with your token max.

There is a way to reel this all in and I have been successful with dozens of companies.

I get paid on the percent of savings and the increase in productive and useful outputs. Or I don’t get paid.

3h2.1K121

REPLIES2

Kevin Sciarra@KevinSciarra

@BrianRoemmele In my company AI usage is "monitored" and part of your performance review. So everyone feels pressure to literally burn tokens to show your "using AI". In 40 years I have never seen anything quite this Stupid.

3h1195

ZeroMazed@ZeroMazed

@BrianRoemmele lol imagine paying for guardrails. my portfolio tracker just shows the carnage, no filters

3h671

Clanker Tees@ClankerTees

@BrianRoemmele heard and understood... what should our next shirt be, commander? https://clankertees.com/?p=tokenmaxxer

3h75

Revo Laition@revolaition

@BrianRoemmele It think there is a lot of fomo at play here. Seen examples where high token usage was praised, with no focus on value produced.

I think many ceos are sobering up, and ai finops will be big business.

3h35

Paula Vazquez@VampireGurlAI

@BrianRoemmele U need to run proper scripts and understand real program language ^or ur gonna run into that frustration! I suggest starting with pi and turtle! 🐢 then go from there

3h32

DC@dct_247

@BrianRoemmele Yep, the usage windows are where the value is, which typically requires a human in the loop. That's where it's most reliable, and doesn't generate slop, when being driven by a subject matter expert

3h24

Cyborg Bear@CyborgBear33

@KevinSciarra @BrianRoemmele The US government does this. Basically, to justify what they receive, they need to spend it all, then they request even more.

3h4

Paisley Circus@PaisleyCircus

@BrianRoemmele Nice value-based pricing

2h3

Revo Laition@revolaition

@KevinSciarra @BrianRoemmele What gets measured, gets managed. Measuring Ai impact is hard, measuring usage is easy.

So they do that. And yes, it’s stupid.

2h2