Thousands of CEOs admit AI had no impact on employment or productivity—and it has economists resurrecting a paradox from 40 years ago

The article argues that AI adoption is widespread but has yet to show up meaningfully in macro productivity data, employment, inflation, or profit margins, despite more than $250 billion in corporate AI investment in 2024. Survey and academic evidence is mixed: some studies show modest productivity gains such as a 1.9% cumulative increase since ChatGPT’s launch, while others estimate only 0.5% productivity growth over the next decade and find usage can become counterproductive with too many tools. The piece is primarily analytical commentary on AI’s uncertain return on investment rather than a market-moving event.

Analysis

The near-term setup is less about AI’s long-run promise and more about a timing mismatch: capex and management attention are being spent now, while operating leverage is leaking into nonproductive experimentation, tool sprawl, and workflow friction. That creates a classic “picks and shovels” versus “ROI proof” split — vendors selling deployment, governance, integration, and security should outperform pure application stories until buyers can demonstrate measurable labor savings. The second-order effect is that AI spend may stay high even if productivity data stay soft, because CFOs are more likely to cut headcount elsewhere first than admit a failed transformation.

For IBM and MAN, the market implication is asymmetric. IBM is better positioned if the cycle shifts from model hype to implementation discipline: enterprise customers will pay for systems integration, orchestration, and compliance when internal adoption is messy, but it is also exposed if clients conclude they can delay projects after failing to see payback. MAN is a more subtle loser: if firms believe AI can replace entry-level tasks, recruiting volumes for junior roles can dip before actual productivity benefits show up, which compresses fee growth with a lag even if headline employment remains stable.

The contrarian view is that weak macro evidence may be a feature, not a bug, because the first productivity gains are likely to be reallocated to leisure, quality, or hidden slack rather than visible GDP. That means consensus may be underestimating the speed at which companies will keep buying AI despite disappointing national statistics. The bigger risk is not a demand collapse, but a second-wave disappointment if boards realize the easy use cases are already saturated and the harder ones require process redesign, not just software licenses.

Key Decisions for Investors

• Maintain a tactical underweight in MAN for the next 1-2 quarters; if AI adoption keeps substituting away from junior hiring, staffing growth can lag sentiment by 1-2 reporting cycles. Risk/reward favors caution until utilization data improve.
• Stay long IBM on a 6-12 month horizon as an implementation-and-governance beneficiary, but hedge with covered calls into strength; upside comes from enterprise replatforming, while downside is slower conversion of AI spend into revenue.
• Pair trade: long IBM / short a basket of high-expectation AI application names with weak monetization visibility. The thesis is that enterprise buyers will pay for integration and control before they pay for discretionary seat expansion.
• Avoid chasing broad AI productivity beneficiaries until there is evidence of sustained operating leverage; use any rally in software multiples to sell into the narrative that macro data will lag by several quarters.
• Optionality trade: buy medium-dated calls on IBM and puts on MAN into the next earnings season if commentary turns to headcount pruning. The best setup is a 2-3 month window where labor data soften before any true productivity inflection is visible.