Uber’s 2026 AI budget was intended to cover employee use of tools like Claude, Cursor, and other tools for the full year. But in just four months, the entire annual budget had been exhausted.

Uber even encouraged employees to use AI as much as possible, creating internal leaderboards ranking teams by their AI usage.

To add insult to injury, the CEO said “it was becoming harder to justify rising token costs without evidence that they were producing more useful features for customers”. (Yahoo)

If a company as technical as Uber couldn't see this coming, it's shows how hard AI cost visibility is for companies and professionals, no matter the industry.

KPMG's recent Global AI Pulse report found that only about a third of companies say they have full visibility into what their AI tools cost day to day. BUT the ones that do are five times more likely to say the spending is actually paying off. (KPMG)

The same math also plays out on a much smaller scale every time you open ChatGPT, Claude, or Gemini.

Whatever tool you use, you've likely run into usage limits. These limits can be especially frustrating if you have a paid subscription (imagine Netflix limiting the number of shows you can watch, no bueno).

The limits are usually dollar amounts behind the scenes, and how fast you hit them depends on a few factors, like which model you're using and how much information that model has to process and generate.

One of the easiest ways to cut that cost is to select the right model for your task.

OpenAI, for example, has the GPT-5.6 family of models consisting of: Sol (most powerful), Terra (for everyday work), and Luna (most lightweight).

Using a more powerful model than what the task requires doesn't get you a higher-quality output AND it burns through more of your usage limit. As the saying goes, bigger doesn’t always mean better.

For any given task or workflow, you should choose the cheapest possible model that can get the job done. Luna, for example, is great for quick answers and basic extractions. Terra is a good "daily driver" for general writing and analysis, and should be your default when you're unsure which model to choose. Sol should be reserved only for problems that genuinely need sustained, complex reasoning.

Another way to cut costs is to look beyond the US models. Chinese AI labs have released free or very cheap models, like DeepSeek and the newer Kimi K3, that perform close to top-tier quality for a fraction of the price. The catch is that they usually take more technical setup to access (safely), so they're a better fit for a team with some engineering support.

In our own industry, most people don't need Sol-level power for their everyday AI use. Summarizing a 40-slide internal presentation, reviewing the first pass of marketing copy or product descriptions, or pulling key data points out of a supplier contract are all tasks a cheaper, faster model can typically handle just as well.

If in doubt, you can always try out a cheaper model first and if the output is not up to par, use a more powerful model instead. This isn’t an exact science and requires some trial and error.

Uber's story is a big, expensive version of a problem any of us can run into at a smaller scale: AI costs will balloon quickly without the right guardrails and habits in place.

Treating efficient AI use as a skill worth building, will become increasingly important as frontier models become more powerful, and expensive to use.

So, the next time you open a chat window, ask yourself: does this task actually need the most powerful model, or just the cheapest one that'll get the job done?

A lighter week on deals, but Estée Lauder more than filled the gap, calling off the sale of Too Faced, Smashbox, and Dr. Jart+ even as its turnaround tab climbed to $1.75B.

Estée Lauder (NYSE: EL) disclosed in a new SEC filing that total restructuring costs under its "Beauty Reimagined" turnaround have risen again to $1.75B before tax, up from a prior estimate.

The added charges bring cumulative job cuts to as many as 10,000 since the program began, though EL still expects $1B to $1.2B in annual gross benefits. Learn more.

Covey, the minimalist skincare brand co-founded in 2021 by model Emily DiDonato and former Google executive Christina Uribe, is shutting down, the founders shared on Instagram.

The DTC brand had raised just an $800K angel round in 2021 and never took on further institutional capital, per Beauty Independent.

Estée Lauder (NYSE: EL) called off the sale of Too Faced, Smashbox, and Dr. Jart+, opting to keep all three and run them on leaner, more entrepreneurial models after private equity buyers passed on a package valued at just $300M to $500M (Too Faced alone was acquired for $1.45B in 2016).

It is ELC's second M&A collapse this year after the scrapped Puig merger; Too Faced will relocate from Los Angeles to New York and Smashbox will operate with a smaller team. Learn more.

General Catalyst, through its Customer Value Fund, committed $1B in non-dilutive growth financing to IM8, the David Beckham co-founded supplement brand owned by Prenetics (NASDAQ: PRE).

The facility funds up to 70% of IM8's marketing spend and is repaid only from the revenue those customer cohorts generate, with no equity issued. Learn more.

Hockey Stick Ventures launched a $3M debut fund backed exclusively by past and present NHL players and the broader hockey community.

Founded by former Numerator sales exec and minor-league player Sean Hershman, the fund will write $50K to $500K checks across 7 to 10 seed and Series A rounds, building on earlier bets including Liquid Death, Recess, The Coconut Cult, and Slate Milk. Learn more.

Sharrp Ventures led a $11.8M Series A in Naturis Cosmetics, the Mumbai-based contract manufacturer's first institutional round, with Mirabilis Investment Trust, Anicut Capital, Niveshaay, and angels joining. Learn more.

Maesa opened applications for the 2027 Maesa Magic Incubator, offering $35K in funding plus mentorship to early-stage founders from underrepresented communities in beauty and wellness.

Now in its fourth year, the program has awarded $315K and launched nine founders; this year's advisory board includes Starface co-founder Brian Bordainick, Topicals CEO Olamide Olowe, Bubble's Shai Eisenman, and leaders from Maesa and Bain Capital. Learn more.

If you think your current processes in any of the following areas could be improved, book a demo and let us show you how our software can help!