By Sarah McIntyre, CEO, Culture CFO
I started listening to Brigid Schulte's Overwork on a new business connection's recommendation a few weeks ago, right around the time HBR recirculated “Collaborative Overload,” a decade-old piece by Rob Cross, Reb Rebele, and Adam Grant. Two sources landing back to back, circling the same question: what happens when the demand on people outpaces their capacity?
Not only did I read these two complementing pieces, but I also attended a few in-person and virtual AI implementation workshops over the past month, among them Oracle's Learning Lab and the World Finance Forum, both held in Philadelphia.
Then I kept seeing the news articles about companies cutting people and crediting AI. But here's the thing: the profits aren't following. The people who remain aren't doing less, they're doing more, for wages that stopped tracking productivity long ago, with fewer support systems than ever underneath them. That's not a productivity story. That's an uncounted cost.
Fewer people doesn't mean less coordination
Cutting headcount doesn't cut the coordination, judgment, and client-facing work still required. Someone has to review the output and personalize it for the end user. That work doesn't disappear, it just lands on fewer people, often the same small group who were already carrying more than their share, but we’ll get to that in a bit.
AI has been the top cited reason for layoffs in the US for five straight months as of July 2026. In January it was 7% of cuts. By summer, over half. Microsoft alone cut 4,800 roles in July 2026 to fund AI infrastructure.
But the savings aren't showing up
MIT's NANDA initiative found that 95% of enterprise AI pilots failed to deliver any measurable financial return, despite 30 to 40 billion dollars in investment.
Gartner found that companies cutting headcount hardest saw no better returns than those cutting the least, sometimes worse. A separate analysis tied only 1% of job losses to actual AI productivity gains. The rest was economic pressure, dressed up as AI strategy.
Goldman Sachs economist Ronnie Walker put it plainly: there's still no meaningful link between AI adoption and productivity at the economy-wide level. Workers report the opposite of the promise, some seeing time spent on certain tasks jump by 346%, because reviewing AI output takes real attention.
Oracle gets cited as proof this works: 25,000 roles cut, net income up 95%. But that's capital reallocation into AI infrastructure sales, not evidence that fewer people are doing more work. Worth knowing the difference before using it as your case.
Economists call this lag the Solow paradox, a new technology's promise outrunning its measurable payoff. It took over a decade for computing's gains to show up in the data. We may be living the same lag now, except companies aren't waiting to find out. They're cutting first.
The part nobody wants to say: the data underneath was never ready
Most organizations run on manual processes, disconnected systems, and data nobody has fully cleaned up. That's normal. It's why implementation work exists.
Point an AI model at that and the failure rate jumps. AI project failure rates rose from 17% to 42% in a year, and 72% of pilots may get shut down over poor data readiness, not technology. MIT Sloan Management Review said it best: automation doesn't fix bad data, it accelerates the impact of it. Gartner puts the cost of poor data quality at 15% of revenue a year, before AI even enters the picture.
Pilots run clean. Production doesn't. Systems that look great in a demo degrade fast against real data, and when problems surface mid-implementation, they add 30% to 60% to the timeline.
This is what gets lost when AI is sold as a replacement instead of a tool. Outcomes depend on data quality, and on whether anyone checks the output before it reaches a client. Skip that verification layer and you haven't removed the risk, you've removed the person who used to catch it.
The capacity gap was already here
A decade ago, Cross, Rebele, and Grant found that the demand for collaboration had ballooned 50% or more over two decades, with people at some companies spending up to 80% of their day in meetings or answering requests. Capacity never scaled to match what was being asked. We all have days that feel jam-packed with meetings, and there's no shortage of memes reminding us how many of them could have been an email.
The interesting thing about the HBR piece is that it found 20% to 35% of an organization's value-added collaboration came from just 3% to 5% of employees. They are your institutional knowledge base, the same small group I mentioned earlier, already carrying more than their share. The researchers also found they're the first to burn out, because the org chart never accounted for what they were actually holding.
Now add AI-driven job loss into that mix, and the capacity gap that was already there keeps widening.
This isn't new, it's just the newest version
Schulte's Overwork makes clear this pattern predates AI. Work used to be compatible with a decent life, one income, real hours. That eroded decades before anyone heard of a large language model. Wages stopped tracking productivity. The supports that used to catch people, stable schedules, pensions, a ceiling on what one person could carry, mostly disappeared.
Worth saying plainly: this is largely a U.S. story. Schulte points to countries like Sweden, where labor law caps the standard work week at 40 hours, mandates real rest between shifts, and backs it with parental leave and a culture that treats leaving on time as competence rather than a lack of commitment. Swedish employees work well below the OECD average and the country still ranks among the most productive in the world. The erosion Schulte traces in the U.S. didn't happen in Sweden because policy and governance decisions there were made in support of the labor force, not around it.
AI didn't create the overload problem. It's the newest version of an old one: ask more of fewer people, and don't rebuild anything to hold them up while they carry it.
The part that should matter to your finance team
Overload isn't a line item until someone quits. Then it's turnover cost, lost institutional knowledge, and often the exact judgment that made that person hard to replace. None of it shows up on a balance sheet in advance.
If you're weighing a headcount cut and calling it an AI strategy, the data doesn't support that bet yet. Before you cut, ask what the people who stay will actually carry, and whether anyone's measured it.
What are you measuring before you make that call?
Sources: Cross, Rebele, and Grant, “Collaborative Overload,” Harvard Business Review, January-February 2016. Brigid Schulte, Overwork: Transforming the Daily Grind in the Quest for a Better Life, 2024. MIT NANDA, “The GenAI Divide: State of AI in Business 2025.” Gartner workforce and AI ROI research, 2026. Challenger, Gray & Christmas monthly job cuts reports, 2026.