Key Takeaways

  • Apollo chief economist Torsten Slok found that non-tech Fortune 500 companies are suffering indigestion over AI capital expenditures, failing to capture immediate returns while Magnificent Seven firms continue to see positive ROI.
  • Goldman Sachs recently clamped down on non-technical workers and traders vibe coding their own software, citing direct risks to firm compliance and actual employee output.
  • Joe Weisenthal explains the messy jobs problem: automating two isolated tasks out of thirty does not eliminate headcount or produce measurable gains.
  • John Coogan notes enterprise buying friction: outside of software engineers, almost no corporate employee can justify a recurring $10,000 monthly AI software budget to their manager.

The Indigestion Problem Beyond the Magnificent Seven

Silicon Valley lives in an echo chamber of massive GPU clusters and rapid model releases. Big Tech companies print cash from their core monopolies, which lets them pour billions into infrastructure and claim immediate returns through cloud sales. For the rest of the corporate world, the arithmetic looks completely different.

On the floor of the New York Stock Exchange, Bloomberg's Joe Weisenthal and host John Coogan dug into recent findings from Apollo economist Torsten Slok. Slok's data shows that Fortune 500 firms outside the Magnificent Seven are hitting a wall. Coogan put it plainly: “His latest is that basically Fortune 500 having a little bit of indigestion around AI spend outside of the Mag7. Mag7 is seeing positive ROI.” (play from 12:30)

When a bank, a logistics company, or a retailer buys enterprise AI seats, the software costs hit operating margins immediately. The revenue offset remains invisible. That gap creates corporate fatigue. Chief financial officers are asking why software line items doubled while headcounts and quarterly output stayed flat.

The Messy Jobs Trap and the Goldman Ban

Why does enterprise software fail to deliver the expected productivity surge? Weisenthal points to job structure. Most corporate work cannot be broken down into clean, isolated prompts.

“We all have, within a job, thirty tasks,” (play from 15:02) Weisenthal argued. “And so you could say, well, AI is really good at this task, but it's not really a replacement for someone's job.” (play from 15:06)

If software speeds up one subtask from twenty minutes to two minutes, the employee does not become ten times faster. They simply check their email or wait on a colleague's review. Worse, handing non-technical teams raw generative software often generates bad internal code. Weisenthal cited a recent move by Wall Street's most visible investment bank: “Goldman is clamping down on its non-technical employees vibe coding, including its traders, which I thought was pretty telling. And it said it was for both risk and productivity.” (play from 14:10)

When traders spend half their day cobbling together scripts instead of watching order flow, trading desks lose money. When non-engineers push unvetted code into production, risk officers panic. The tool ends up bleeding hours rather than saving them.

Even enterprise tooling budgets have hit a hard ceiling. “Right now there is no one outside of an engineer that can go to their boss and say, 'I want to spend $10,000 a month on AI.' That is pretty hard,” (play from 13:58) Coogan noted. If the purchase requires five approval signatures and cannot prove it replaces head count, the deal dies in procurement.

Weisenthal compared the current skepticism to historical technology adoption: “What will be the productivity effects of AI? It is like: what were the productivity effects of the internet? It is hard to see. You don't see them in the chart, even though, what were the productivity effects of the mobile phone?” (play from 9:30)

Macroeconomic charts rarely register new technology cycles as clean vertical spikes. For early-stage builders, selling general productivity to traditional enterprises is an uphill battle. If your product only shaves five minutes off a secondary task, buyers will freeze your contract the moment budgets tighten.

What to Do With This

Audit your product pitch or internal tool requests this week. If you are selling software that promises vague productivity across twenty different tasks, cut that positioning immediately. Rebuild your sales deck around a single, auditable workflow that either completely eliminates a contracted vendor cost or automates an entire end-to-end operation from intake to delivery.