Table of contents
In his latest Daily Spark, Apollo chief economist Torsten Slok argues that AI has yet to pay off outside the tech sector. Revenue per worker and profit margins are climbing for the “Mag 7” tech giants but staying flat for the rest of the S&P 500. Looking ahead, he’s skeptical that mainstream companies will close the gap, because business decisions depend on judgment, and AI doesn’t understand things like accountability, trust, culture, and unwritten rules.
His analysis of AI’s impact so far is excellent, as always. I do have a different take on his go-forward predictions, however.
The facts are clear: AI is generating scattered positive impact, combined with established and sometimes significant cost. Nearly 9,000 executives told Bloom and co-authors (NBER, October 2026) that AI raised productivity 0.2% a year. As Bank of America reported on October 13, “We’re about 130 different projects implemented. All of them have a payback … None of them are huge contributors.”
I see three drivers of this lack of impact: two costs and one unrealized benefit.
- Swamping: Employees can process information slightly faster with AI, but they generate and share content much faster. This adds to a growing pile that slows down decision making.
- Re-asking: AI lowers the cost of colleagues generating their own answers to common questions.
- Coordination: AI’s capabilities, proven at the individual level, can provide cost-efficient support for organizational alignment, solving the “broken telephone” problem. But this kind of organizational setup is still new.
Alignment agents are a potential solution to all three. Alignment agents keep the organization’s information coordinated, set strategy from it, and keep execution aligned to it.
In particular, AI doesn’t need to understand “accountability, trust, culture, unwritten rules” itself. It needs to show leaders where strategy and execution have veered apart, with high enough accuracy to be useful as decision support.
And unlike electricity and IT, which Slok noted on October 5 took a decade or more to show up in the numbers, an alignment agent can be set up in days and needs no organizational redesign. So I expect a faster payoff curve. Ethan Mollick recently described AI helping with this coordination challenge with quick setup. And Bonney et al.’s Census Bureau study of 111,300 observations points the same way, without proving causation: breadth of organization-wide AI deployment tracks performance, while worker-level task use does not.
