A Poem For FinOps

For next week’s monthly Show Me Something Cool With AI, I issued an optional challenge to get AI to write a poem that sounds like you in the FEWEST number of words/prompts as possible.

This challenge is both a curiosity of mine and my forecast where AI Fluency is headed next.

First, the forecast: people are using AI. According to Pew Research, about half of Americans are using chatbots in 2026, up from a third in 2024 and 23% in 2023. You don’t need much high school math to see the trend.

In both the learning and drug dealing world, basic adoption is the first hurdle. Give newcomers a free taste. Get them hooked.

The next phase of change is using AI effectively. This is what your LinkedIn feed and AI hypesters are promoting:

• How to prompt AI better
• How to do more complex tasks with AI
• How to harness advanced AI features
• How to build AI-powered thingamajigs

A lot of this effectiveness comes from quantity. More examples, more feedback, more back-and-forth, more text, more tokens. This technology was built by hoodwinking the entire internet, so no surprise that AI gets smarter when you fuel your AI tools with lots of context.

I’ll refer to this as the 5xMore habit.

Once humans have made AI their go-to resource and they learn how to do more than magic tricks or rewriting emails, 5xMore will be treated as the panacea to ineffective AI outputs.

But the direction my poem prompt is leading is the forecast every CFO, business owner, and climate warrior will be talking about in the next few years:

Efficiency, The Next Big AI Trend

AI use in 2026 is measured by tokens. I hope that sentence gets outdated soon, but for now, that’s the imprecise way we measure how many energy-intensive server clusters get spun up when you want an AI-generated recipe that uses whatever is left in your fridge. If you use a frontier reasoning model, your recipe request could cost 20 times more tokens (Artificial Analysis) to get the same meatloaf as another, more efficient model.

Not only that, but when you are doing longer-form work with AI, the entire transcript getting re-fed back into the data center with every “Oh yeah, I forgot, make it gluten-free please” request causes exponential token growth.

More tokens, more servers humming, more energy consumption, more data centers in your backyard, more climate change. Shoving 5xMore into a chatbox is being paid, somewhere, and the cost is invisible to you while you make passable meatloaf.

The large frontier AI labs are excellent drug dealers. They’ve obfuscated token costs into a flat subscription. Most users won’t think about token spend for $20/mo as they happily keep entrenching their AI habits without the friction of weighing each 13-cent request. Only when you’ve exceeded a plan’s limit do they start charging a la carte, which has created an ignoble competition amongst superusers that is being called tokenmaxxing.

Yeah, there are Silicon Valley nerds measuring the size of their 🍆 by how expensive their unbridled AI use ends up being. Jensen Huang kicked off tokenmaxxing culture in March on the All-In podcast: if his $500K engineer didn't consume $250K in tokens, he'd be "deeply alarmed.”​

However, tokenmaxxing is likely just a transition blip in the broader adoption story. You might be able to see the future, now:

1. People start using AI, widely.
2. The more they use AI, the better they get at using AI effectively.
3. Without token cost friction, effective will mean use more tokens
4. Everyone’s habits will be to 5xMore.

Congratulations will go to the winner of the AI race: you’ve gotten an entire society trained to spend tokens like junkies.

And what do you do to junkies? Raise the price.

And who will witness our incredible token spend to feed our addiction to technological magic as it becomes a worryingly large expense?

The CFO.

Our hero of ledgers and capital expenditures. The authority on “this is a bad way to spend money.” The only mortal among us with the gravitas and data to be the canary that unbridled AI use is financial suicide.

This has already happened.

Efficiency, The Next Big AI Trend Now On Every CFO’s Radar

Up until recently, CFOs have trusted the promises by everyone else that rising AI expenses (licenses, tokens, trainings, etc.) delivers a return on investment but that trust has already been shattered in less than five months:

• ​Meta ran an internal leaderboard called "Claudeonomics" where employees competed for status as a "Session Immortal" or "Token Legend". If being called a Token Legend by Zuckerberg isn’t 🍆 measuring, I don’t know what is.
• ​Uber burned through its entire 2026 AI coding budget in four months and had to slap a $1,500 monthly cap on each employee.​
• ​Gartner predicts AI coding costs will surpass the average developer's salary by 2028.​
• ​Chamath Palihapitiya told CNBC ten days ago, “I suspect what’ll happen is one day you’re going to have a miss, and EPS will be off by a few pennies, and the CEO will say to the CFO, ‘What happened?’”​

I’m not going so far to say that our AI addictions will drive us to financial insolvency, but it has proven to be a tough challenge to retrain an entire society (including the tokenmaxxing cults) to use AI efficiently.

Attempts to solve inefficient token spend with technical fixes get stymied by the very human desire to use the best, fastest, smartest AI model you can grab onto. Benioff says Salesforce's Anthropic bill will hit about $300 million this year and wishes for a "smart router" to send cheap queries to cheap models, and if Benioff says it’s a wish, that means it ain’t a reality.

Or worse, CFOs demand token austerity with the not-so-helpful guidance to, “Be mindful of your usage.” As if you’re NOT going to drive the sports car that’s sitting in your chat window simply because you deeply care about your company’s bottom line.

Nobody is teaching the correction.

Bad habits are much harder to fix than making good ones in the first place. Instead, let’s make a more beautiful habit that will keep our CFOs happy.

All The Fun’s in How You Say a Thing

In my undergrad at Brandeis, I was a quintessential liberal arts kid who took so many English classes for fun that I accidentally minored in it. Fun for a learning nerd looks like countless drafts where- 3/4s of the way through a story that took me a weekend to write- a red pen simply said “STORY STARTS HERE.”

Often, it was right, and that haunts me, joyfully. For me, reading, crafting, and discussing stories has proven to be both an important muscle for clarity and a meaningful connection to humanity.

And it is the humble poem- the haiku, the sonnet, the opus- where you have to work hard to find just the right word to convey a specific and vital human truth. To write GOOD poetry, you must believe that “Omit Needless Words” is a fundamental truth for writing, basically a catechism. And the “Write From Experience” advice that every novice poet receives is the training wheels to learn how to tell authentic and compelling stories.

Holy smokes, that catechism and those training wheels have real financial benefits when it comes to token spend. You don’t need a novel’s word count for an AI to mimic a specific writing style, and you desperately need a human with lived experiences to identify that something smells like slop before your customers notice.

These are expensive mistakes stemming from sloppy habits nurtured by Big AI, your corporate token dealer. But what if poetry, a human taste skill that we have been practicing since before the Odyssey, could actually deliver the efficiency that CFOs and our climate need?

So here’s my curiosity:

Come next week to SMSCWAI and find out.

(Oh, and if you are willing try a little writing assignment to improve your EFFECTIVE AI skills before trying some EFFICIENT prompting, follow this guide to create a Style Fingerprint and figure out what YOUR unique voice looks like)

A TL;DR from the CRO

My overgrown snoot
can sniff out extraneous
tokens and tuna.

-Roman Noodles, Chief Ruff Officer

A short TL;DR from a better writer:

I stumbled across a great little piece by Ricardo López Cordero about how the Mexican government tries to make policy happen by simply adding it to their constitution.

It doesn’t really work.

The AI governance take: If you think drafting AI policy into an AUP is gonna make people give up their personal Claude account, buena suerte, amigo!

I crashed a conference's afterparty and after-afterparty while on a work trip. Amidst pickleback shots and kareoke, complete strangers were thrilled to chat with me about AI and its impact on the travel industry. All I had to say was, "I do couple's counseling for organizations and AI," and they just unloaded all their hopes, fears, and examples of how their bosses are embarrassingly behind the times.

I'm sure the picklebacks played a part in those confessions, but I wonder if I should just go to conferences and set up a psychiatrist booth like Lucy from Peanuts where people can just vent.

How far can five cents get you in API fees? Depends on how elegantly the question is phrased.

The doctor is in,

Dan from Learn to Scale​

​Opt-out from the newsletter | Unsubscribe from all emails | Update your Preferences | www.learntoscale.us, Boston, MA 02119

​PS. Poetry hits hardest when it slams​