Two different perspectives on the impact of AI: the workforce versus the investor. Who’s right?

There is a growing gap between two stories about AI and work, and they point in somewhat opposite directions.

Story one: the one most professionals tell themselves at the coffee machine. “As long as I learn to work with AI, I’ll be fine. My job won’t be overtaken. I’m on the safe side.” It is comforting, and not entirely wrong: LLMs are not perfect (yet), companies are complex, and transformation does not happen overnight.

Story two: the one investors are betting hundreds of billions on. AI is not eating the Software-as-a-Service (SaaS) market. It is eating the services market. And the services market is, almost by definition, white-collar jobs. So do not deliver a tool to help the accountant closing the books. Be the accountant. And close the books.

Both stories can be partly true. But the financial bets are large enough that it is worth pausing to understand the economic consequences of each.

I. THE INVESTOR THESIS: SERVICES ARE THE NEW SOFTWARE

Start with the investor thesis. In March 2026, Sequoia’s Julien Bek published “Services: The New Software.” The argument is blunt: the next trillion-dollar company will not sell software. It will sell the work itself, powered by AI, delivered as an outcome.

Why? Because for every dollar enterprises spend on software, they spend roughly six on services. The total addressable market is no longer “the software budget.” It is all labour spent in a category, insourced and outsourced combined.

That reframing is now embedded across venture capital. Recent pitch decks have stopped sizing the SaaS line item and started sizing the salary line item of the category they want to replace. The Total Addressable Market (TAM) for an AI legal-services company is no longer “the legal-tech market.” It is the wage bill of paralegals and junior associates. Sequoia puts numbers on the targets: recruiting and staffing ($200B+), IT managed services ($100B+), medical billing ($50–80B), accounting and audit ($50–80B), transactional legal work ($20–25B).

The proof point investors point to: founders running real services businesses with almost no people. These companies do not talk about onboarding flows or pretty User Interfaces. They focus on getting the work done and taking end responsibility. The thesis, distilled: do not sell tools to humans who do the work. Be the work.

Corporate boardrooms are following suit, with large lay-offs and restructurings, often with AI cited as the rationale. The story is more nuanced than that. In many cases real productivity gains remain unproven and therefore a thesis. The cuts are also funding something else. For instance, the capex bills coming due. Microsoft eliminated more than 15,000 roles in 2025 while committing over $80B to AI infrastructure. Amazon cut roughly 14,000 manager roles. Meta let 3,600 people go early in 2025 and another 8,000 in May 2026. Across the tech industry, an estimated 55,000 layoffs in 2025 were tied directly to AI.

Or take Workday, a roughly $8B-revenue HR/finance SaaS. In February 2025 it cut 1,750 jobs – about 8.5% of its workforce. What’s behind it: the stock dropped more 40%+ in less than a year as investors fear that AI is eating SaaS. Revenue growth and margin growth were actually “healthy” (note: I am not a financial expert). But the MT had to turn the ship around regarding the investor sentiment and doing more with less people is a critical surviving strategy they are executing on today. 

Overall, first results show that companies – also Workday – see revenue per employee rising. Margins per employee will follow. And so will the money flowing out of the business to shareholders and investors. In the long term: the workforce is losing, shareholders and US big tech are winning.

II. THE WHITE-COLLAR THESIS: AI WILL MAKE ME MORE PRODUCTIVE

How do workers themselves feel? CBS data from the Netherlands is striking: 75% of Dutch adults expect AI to cause jobs to disappear. When zooming into their own job, that fear drops. Among working people, 41% think AI can partly replace their work, and only 4% think it can fully replace it. Similarly, most experience AI as a productivity gain, but not as a reason to expect labour shortages to be solved any time soon.

Interestingly, the people who use AI most are also the most worried. 56% of active AI users think their work could be partially or fully done by AI, versus 37% of non-users. Experience, in this case, does not breed comfort.

Anthropic’s recent exposure mapping, using real Claude usage data, finds the highest-exposure roles are in tech, finance, law, consulting and administration. Dario Amodei has been blunt about it: he expects AI could eliminate nearly half of all entry-level white-collar jobs in those categories. Roughly 30% of workers – cooks, mechanics, electricians, nurses – have essentially zero LLM exposure.

So why, then, has the wave not crashed harder onto knowledge work? Several reasons. AI output is still inconsistent: hallucinations, edge cases, and weak handling of ambiguity make full autonomy risky. Real work happens inside organisations with messy legacy systems, fragmented data, and internal politics that no model is going to refactor. Regulations in finance, healthcare and law require a human in the loop; not just for ethics, but for liability. And transitions take time: rolling out AI through a company demands change management, retraining and trust, all of which move at the speed of humans, not models. None of these reasons are permanent. But they buy years, not months.

III. WHICH STORY WILL DOMINATE: AND WHY THAT MATTERS

So we have two mental models living side by side. At the desk: “AI makes me faster.” Versus in the boardroom and at shareholder meetings: “we can resize this team.” The two theses have very different economic consequences.

The first model is about productivity. That matters. European productivity has been flatlining for years, and AI can be a real boost. Especially for a country like the Netherlands, whose economy leans heavily on services.

The second model has a sharper edge. The same services market that powers Dutch GDP can be automated away by AI. The impact is multifold, but the part that doesn’t get said enough on LinkedIn is this: when we replace services with AI, we are not just changing how the work gets done. We are moving a line on the P&L from “salaries” to “AI tokens.” And those tokens are sold, overwhelmingly, by large American companies: Anthropic, Microsoft, Google, Amazon and Meta.

This shift is accelerating. In 2026, the size of your “model bill” has become a corporate status symbol. A flex about who is leaning in hardest. Underneath it, the largest US tech companies are projected to spend ~$725B on AI capex in 2026, up 77% year-on-year, with Wall Street estimates climbing past $1T by 2027. 

What does that mean concretely? A euro that used to go to a paralegal in Utrecht, an analyst in Rotterdam or a support agent in Groningen – a euro that paid a mortgage, sat in a local café, funded a school sports club – increasingly becomes a euro that goes to a hyperscaler’s data centre bill in Virginia.

IV. WHAT I’D WATCH FOR