FORMA's Five Things.

FORMA Programs

Build With AI is coming.
Tell us what you want to learn first.

Claude Code. Cowork. AI Dashboards. Deep Research. We are designing the next session around what you actually want to get your hands on. Click below, pick your topics, and you will be the first to know when we announce the date.

Participants in FORMA's AI Fluency program report an average 25% increase in AI confidence after a single session.

Jensen Said Spend More.
The CFOs Said No.

Back in March, Nvidia's Jensen Huang dropped a line that got everyone talking. His view: if you pay an engineer $500,000 a year and they only spend $5,000 on AI tokens, you should be alarmed. His benchmark was half their salary, minimum, in compute. If they fell short of that, he said, he would "go ape something else."

The CFOs had a different reaction.

By April, companies were calling in a panic. Uber burned through its entire 2026 AI coding budget in four months. One company reportedly ran up a $500 million Claude bill after forgetting to set employee usage limits. The FinOps Foundation started fielding calls from finance teams already 3x over their annual token budgets, and it was not even summer.

Even Sam Altman acknowledged it. Token overspending, he said, had officially become "a meme." The era of use AI at any cost ran headfirst into the era of who approved this invoice. The reckoning is here, and a lot of companies are now figuring out the difference between AI adoption and AI discipline.

Read more →

Over 1,000 Court Cases Have AI Errors.
And Those Are Just the Ones They Caught.

Did you know the court system has now identified over 1,000 cases where AI-generated hallucinations made it into official legal filings? Fake case citations. Invented legal standards. Quotes that do not exist in the cases they are supposedly pulled from. Over one thousand. And that number only counts the ones a judge actually caught. The ones that slipped through? Nobody knows.

The pattern keeps repeating. A firm submits a filing with fabricated citations, gets called out, apologizes to the judge, promises new AI policies, then does it again in a different court months later. Sullivan and Cromwell, an Am Law 100 firm, recently had to file an emergency letter cataloging roughly 40 corrections across multiple filings. The letter itself became a story.

The takeaway is not that lawyers are careless. It is that AI output looks exactly like real output. Same confident tone. Same citation format. Same plausible-sounding case names. The model does not flag what it invented, because it does not know it invented anything. If you are not verifying, you are trusting. And in a courtroom, that is an expensive habit.

Read more →

Anthropic Just Filed for an IPO.
Now Comes the Part Where Everyone
Finds Out What Is Real.

Anthropic confidentially filed its IPO prospectus with the SEC last week, becoming the first major AI lab to make that move ahead of OpenAI. The company is riding serious momentum: a $965 billion valuation, a reported $47 billion revenue run rate, and the distinction of having eclipsed OpenAI in valuation for the first time.

That is the headline. Here is the question underneath it.

Going public means opening the books. Audited financials. Public scrutiny. Real numbers attached to real claims. The AI industry has been running on narrative, investor conviction, and deferred accountability. The IPO window forces a different conversation: not what these companies could be worth, but what they actually are worth right now. Experts are already drawing comparisons to the dot-com era. One analyst framed it plainly: this either becomes the most consequential IPO cycle since the late nineties, or the most expensive lesson in narrative-versus-fundamentals that public markets have ever taught. Either way, we are about to find out.

Read more →

The People Who Swore They Would
Never Use AI Are Using AI.

For the past two years, the loudest voices against AI in Hollywood belonged to exactly the kind of filmmakers whose opinions move the industry. The strikes, the open letters, the red lines. The creative community had a unified message: not us.

That wall is cracking, and the names doing the cracking are not small ones.

Martin Scorsese just signed on as an adviser to AI image company Black Forest Labs, and has already used their tools to storyboard his next film. Steven Soderbergh used AI to generate whole sequences in a recent documentary. James Cameron sits on the board of Stability AI. Peter Jackson called it "a special effect." Ben Affleck quietly built an AI company and sold it to Netflix. Doug Liman shot an entire film on AI-generated stages.

These are not tech evangelists. These are the people who built the argument for human craft. And they are using the tools.

What they are not doing is letting AI make the creative decisions. That is the distinction worth watching. The directors embracing AI are using it to move faster, visualize earlier, and spend their energy where it matters most. That is not a surrender. That might actually be the model.

Read more →

The Instructions Nobody Told You About

Every time you open a chat window with an AI, you think you are starting from zero. A blank slate. Just you and the model.

You are not.

Before you type a single word, there is already a layer of instructions running underneath. It tells the model who it is, what it can and cannot say, what tone to use, what to refuse, and who it is talking to. That layer is called a system prompt.

You have felt it without knowing what it was. A tutoring bot that will not just write your essay and instead asks guiding questions. A customer service AI that will not acknowledge a competitor exists. A productivity tool that gently suggests you stop working and get some sleep. None of that behavior came from the model itself. Someone wrote instructions telling it to act that way before you ever showed up.

Here is what that means practically: when an AI declines to help you, hedges on an answer, or seems to be steering around a topic, it may not be a limitation of the model. It might be a deliberate instruction from whoever built that product. The model is just following orders you were never shown.

Once you know that, you stop blaming the AI for being unhelpful. You start asking a better question: whose instructions am I actually working inside right now?

FORMA Tools

Not sure which AI model is actually right for your work?

Six questions. A personalized model recommendation. No benchmark tables, no guesswork. Just a clear answer based on how you actually work.