A year has passed. A large new study came out from NBER, the Bank of England, and the Bundesbank — and guess what? The numbers haven't changed at all ¯\_(ツ)_/¯
Analysts surveyed about 6,000 executives from the largest US and European corporations (CEOs, CFOs, and other important people with three-letter titles). The enterprise picture right now looks like this:
— 70% have somehow dragged AI into their business processes
— Only 20% say neural networks actually improved employee efficiency
— And just 11% of companies showed tangible labor productivity gains (revenue per employee went up)
— 1.5 hours per week is how much time executives actually spend in AI tools
The problem is that handing out corporate subscriptions to ChatGPT and actually rebuilding dusty pipelines are fundamentally different tasks. It's one thing to run pretty benchmarks and evals, and quite another to make AI work without hallucinations on your legacy database — and in a way that makes money, not just saves half an hour on writing emails.
The funniest part of all this is the wild contrast between harsh corporate reality and what startup founders are selling us. Microsoft confidently predicts that by the end of 2027 AI will do all white-collar work. And Anthropic's CEO last year said neural networks would fully replace programmers and write 100% of code by March 2026.
I remind you that March 2026 arrives in literally a couple of weeks.
No surprise that regulators don't share the techno-optimism — the Bank of England itself urges investors to prepare for a deep recession if the current AI bubble suddenly pops. In short, stretching pretty Twitter demos onto real boring business processes turned out to be much harder than just pouring billions into data centers.