Calm down about AI efficacy pilots and hiring pauses
Welcome to Cautious Optimism, a newsletter on tech, business, and power.
Thursday! Today we have two longer sections, the first on market upset over the efficacy of AI pilots. And then a note on the White House-commanded halt on solar projects and the startups that could find themselves in trouble due to the policy change. To work! — Alex
📈 Trending Up: Nuclear tensions? … fintech IPOs … phones that cost more than laptops … DeepSeek, after its V3.1 model excels … Chinese AI more generally … the price of bitcoin? … Anthropic’s revenue … self-driving IPOs …
📉 Trending Down: Car dealerships … science at the Federal level … the HPE-Juniper merger? … Walmart, after earnings … tourism … privacy … gaming affordability …
Quick Hits
OpenAI reaches $12B run rate: Reporting that July was its first month with $1 billion or more in revenue, OpenAI has scaled 20% from its June-era $10 billion run rate. That’s pretty damn rapid growth for a company of OpenAI’s scale. At current pace, OpenAI could close the year at around a $17 billion run rate, though our confidence in that projection is modest.
Keep an eye on age verification: Age verification laws in the UK, Florida, Texas, and other areas are taking hold. While noble in theory — protect kids from seeing media that isn’t age-appropriate! — age-gating much of the Internet leads to the blocking of all sorts of non-explicit material. Proponents of the laws like Russel Vought (previously Project 2025, currently director of the OMB) of have been candid that their goal is to get adult content off the Internet more generally, using age gating as a method to achieve the aim. The Supreme Court has come along, too. What counts as obscene is flexible, and could include in more conservative states blocks on things like LGBTQ information, criticism of select nation states, and the like. And an end to anonymity online. And currently age verification is winning.
Calm down
Yesterday, CO noted that an MIT report on the efficacy of AI pilots in corproate settings, with the headline metric snagging all sorts of media coverage (and gloating). It’s worth taking a moment to read the report in question to determine if “95% of corporate genAI projects fall short of success,” as ComputerWorld put it.
First, the report does not argue that individual use of AI tools in a corporate setting generates little or no value. The opposite, in fact. The report — which you can read here — describes an AI divide:
While official enterprise initiatives remain stuck […] employees are already crossing it through personal AI tools. This "shadow AI" often delivers better ROI than formal initiatives and reveals what actually works for bridging the divide.
In simple terms, a large share of total AI usage inside of corporations is being driven by individual actors, demonstrating, the report argues, that “individuals can successfully cross the GenAI Divide when given access to flexible, responsive tools.”
What the report does say is that:
[E]nterprise grade systems, custom or vendor-sold, are being quietly rejected. Sixty percent of organizations evaluated such tools, but only 20 percent reached pilot stage and just 5 percent reached production.
This means that many companies building on top of AI models struggle to develop enterprise software that can successfully navigate the pilot-to-production pipeline and reach production. What’s holding them back?