Overview of Why Businesses Can’t Budget for AI
This episode of WSJ What's News covers the day’s biggest headlines before digging into a practical business problem: companies are struggling to forecast what AI will actually cost. The central theme is that AI usage is highly variable, model choice matters a lot, and “cheap” tools can end up being more expensive in token usage than higher-end ones. The episode also looks at how AI is subtly changing the way people communicate at work.
Top News of the Day
Global and legal news
- Navi Pillay, former U.N. Human Rights Chief and current International Court of Justice judge, won the Nobel Peace Prize.
- The Nobel Committee highlighted her long record on international cooperation, justice, and war-crimes accountability.
Fort Hood execution plan
- The Pentagon said it plans to live stream the firing squad execution of Nidal Hasan, the former Army officer who killed 13 people in the 2009 Fort Hood shooting.
- Defense Secretary Pete Hegseth called the decision a “no-brainer,” saying the public should see the consequences of such crimes.
- It would be the first U.S. military execution in more than 65 years, scheduled for December 3.
Immigration and protests in New York
- An ICE officer shot and injured a man in New York City, sparking protests and renewed calls for ICE to leave the city.
- Mayor Zohran Mamdani said the man was in a car with a five-year-old child, who was unharmed.
- DHS said the shooting happened during a targeted arrest operation, but gave few details.
Markets and tech
- AI chip stocks such as NVIDIA and Micron fell after reports suggested OpenAI’s revenue may have come in below earlier signals.
- The report was later questioned, and some of those stocks recovered.
- A cloud computing company backed by NVIDIA scrapped its IPO, underscoring investor concerns about AI valuations and infrastructure spending.
- SpaceX shook the wireless industry after buying spectrum licenses, raising the possibility of a future challenge to AT&T, Verizon, and T-Mobile.
Why Businesses Can’t Budget for AI
The core problem: AI usage is hard to predict
- Businesses are trying to budget for AI in an environment where usage can change week to week.
- A major issue is token consumption: employees can choose different models and different levels of “thinking,” which dramatically affects cost.
- In a recent study, only 11% of organizations could budget accurately for AI use.
Cheap models can be surprisingly expensive
- A study involving researchers from Stanford, Carnegie Mellon, UC Berkeley, and Microsoft found that in 32% of cases, lower-cost models ended up costing more in tokens than pricier ones.
- The reason: the cheaper model may be the wrong tool for the task, leading to more back-and-forth, more steps, and more wasted spending.
A vivid example from WSJ’s testing
- The Journal compared two Google models on the same task:
- Gemini 3.1 Pro solved the task efficiently.
- Gemini 3 Flash failed, but used roughly 10 times more tokens.
- In one example, the cheaper model:
- Hit bot-detection issues,
- Repeatedly retried the task,
- Even tried other chatbots and Bing,
- Ultimately failed while still generating a bill.
What businesses can do
- Experts recommend:
- Training employees to choose the right model for the right task,
- Setting token limits or usage controls,
- Accepting that AI costs will remain imperfectly forecastable as the technology evolves.
How AI Is Changing Workplace Language
“Prompt speak” is seeping into everyday communication
- Journal reporter Catherine Bindley described how people are beginning to talk like chatbots and prompts:
- Using contrast framing (“It’s not this, it’s that”),
- Adopting short, direct “caveman speak” like “why no bullets?”
- These habits are showing up in:
- Team meetings,
- Group chats,
- Casual office conversation.
The downside: sounding generic
- While prompt-like phrasing may work well with AI, it can make people sound:
- Less distinctive,
- More formulaic,
- And sometimes less engaging in human settings.
Key Takeaways
- AI is difficult to budget for because actual usage depends on model choice, task complexity, and how people interact with the tool.
- Lower-cost AI tools are not always cheaper overall if they require many more tokens to reach a result—or never solve the task at all.
- Businesses need to move from blanket AI adoption to smarter governance and model selection.
- Beyond cost, AI is also changing workplace culture and language, influencing how people write and speak.