Otis CEO Judy Marks said something fascinating a few weeks back.
Her company employs roughly 45,000 elevator mechanics. The top earners make close to $159,000. Most stay in the trade their entire careers. And she says Otis can’t hire them fast enough.
In the age of AI, elevator mechanics are one of the most stable, well-paid, and durable careers in America.
A pattern emerges across domains
The same hiring picture is showing up across very different parts of the economy.
Jensen Huang has noted that the current AI data center build-out is the largest infrastructure build in human history. This will need skilled trades workers by the hundreds of thousands.
This means electricians, plumbers, HVAC technicians, welders become highly sought after. Data-center electricians in Texas earn into the mid-six figures, and trade-publication reporting puts the upper end of those packages at $240,000 to $280,000.
New JLL skilled-trades research quantifies just how lopsided this has become: nearly 600,000 jobs were posted for major skilled trades roles in the U.S. last year, while only about 150,000 new workers entered through apprenticeship programs. JLL projects roughly 2.1 million trades positions could go unfilled by 2030.
What’s even more interesting is, the same pattern is showing up across knowledge work like software engineering.
In May 2026, Anthropic, Blackstone, Hellman & Friedman, and Goldman Sachs launched a roughly $1.5 billion enterprise AI services firm whose entire model is embedding Anthropic engineers inside client organizations.
Every major frontier lab is racing to scale these forward-deployed teams and every major consulting firm is wiring deep partnerships around them. The job-board signal is clear: Postings for forward-deployed and applied AI roles have grown faster than almost any other technical category to 5,230% above January 2025 levels.
It shows up in storytelling. The most successful newsletter operators out-earn entire content teams at legacy publications. Top founder ghostwriters command six-figure annual retainers because they live inside a specific founder’s voice in a way no model can replicate.
Brand storytellers at the high end of the market are commanding monthly retainers that exceed full-time content-team payroll, even as the generic content market collapses around them.
Google, Microsoft, Vanta and others are paying north of $200K for brand storytellers.
It shows up in AI deployment. OpenAI countered Anthropic’s May enterprise AI launch with a $4 billion “Deployment Company,” Amazon Web Services committed $1 billion to a new, large-scale deployment team, and Microsoft introduced a similar $2.5 billion “Frontier Company” initiative.
Every major frontier lab and hyperscaler is racing to scale these forward-deployed engineers (FDEs) and every major consulting firm is wiring deep partnerships around them. The job-board signal is clear: Postings for forward-deployed and applied AI roles have grown faster than almost any other technical category to 5,230% above January 2025 levels.
It shows up in cybersecurity. ISC2’s 2025 workforce study puts the global gap at roughly 4.8 million unfilled roles. The industry faces a paradox: the global workforce is at a record high of 5.5 million, but supply cannot keep pace with rapidly accelerating threats. This will expand further with the potential release of Claude Mythos and “Mythos-tier” models that ramp up cybersecurity threats.
It shows up in nursing. The WHO projects an 11 million healthcare worker shortage by 2030 and U.S. health systems report record open postings. It shows up in industrial maintenance, surgical specialization, trial law, and customer success at companies with complex enterprise contracts.
What do they all have in common? These are all what I’ll call last-mile workers. They work where a system meets reality. They translate between something abstract and the specific context where the work has to land. This may be a model, a building, a contract, an audience, or an organization’s messy systems/data.
The elevator mechanic works at the last mile of mechanical engineering. The forward-deployed engineer works at the last mile of frontier AI. The embedded storyteller works at the last mile of a brand’s voice. The data-center electrician works at the last mile of the AI buildout. The nurse works at the last mile of medicine.
The work cannot be done abstractly, it cannot be done generically and it is structurally hard for current AI to fully automate.
So why does AI struggle at the “Last Mile”
AI is rapidly absorbing templated execution. What it struggles with is presence. Three things about presence are hard to replicate.
1. Context lives in people, NOT in prompts.
Most of what matters in a real organization is unwritten. It’s in someone’s head. It’s in last quarter’s failed initiative no one wants to talk about. It’s in the political tension between two VPs. It’s in the founder’s three-AM intuition about what the company is actually for. The person holding this context usually doesn’t know they’re holding it, which is why it can’t be transferred to a model. You only get it by being in the room when something goes wrong.
2. The work changes the reality it operates in.
A discrete task leaves the world unchanged. Last-mile work is a feedback loop. Each action changes the conditions under which the next action will happen. The deployment redesigns the workflow, which changes the team, which changes what next quarter’s deployment needs to look like. It is recursive in a way that AI models cannot fully capture today.
3. Accountability cannot be delegated.
When the elevator breaks again, the mechanic gets the call. When the campaign falls flat, the storyteller must face the founder. This accountability does work the output itself doesn’t do. It changes how decisions get made upstream because the person making them knows there’s someone real on the other end who will have to answer for the result. No buyer treats AI output with the same weight as work produced by someone with reputational risk on the line.
The last-mile worker offers a vantage point AI struggles to occupy.
What’s Actually Happening to Knowledge Work
Every knowledge profession is splitting in two.
The templated middle: generic copy, basic legal review, deck production, production agent development, literature summaries, first-draft analysis, is getting eaten by AI quickly. This part of the market is contracting, and the contraction will accelerate.
The last mile: the embedded, specific, contextual, accountable work where judgment compounds are getting scarcer AND more valuable.
The data is starting to show it:
- Templated middle is getting automated: Extensive labor market analyses, including landmark studies published by institutions like MIT, confirm measurable declines in job postings for highly automated creative and technical fields since the release of generative AI tools.
- Demand for AI fluency increasing: McKinsey’s State of Organizations research finds that demand for “AI fluency,” the ability to actually deploy these tools inside real work has grown nearly sevenfold in two years, faster than any other skill category.
- Value AI Fluency is rising: PwC’s Global AI Jobs Barometer shows that even in the most highly automatable jobs, AI can make people MORE valuable, not less.
The market is deleting jobs where your value was doing the task and rewarding jobs where your value is working at the last mile. The solution requires figuring out where AI plugs into your specific reality and owning the outcome.
No profession is exempt. This is happening to lawyers, designers, writers, analysts, marketers, engineers, consultants, researchers, doctors, teachers, and operators at the same time.
You Need a New Posture
You don’t need to quit and become a forward-deployed engineer at Anthropic. You don’t need an apprenticeship at Otis. You don’t need to start a newsletter or get a welding certification. You need to stop doing the templated middle of your current work and start working at the last mile of it.
In every job, in every profession, there’s a templated layer and a last-mile layer. The templated layer is the part you could write a process doc for. The last-mile layer is the part that requires being inside the specific context. This includes your customers, your colleagues, your founder, your codebase, your industry’s quirks and owning what happens.
Most knowledge workers spend 80% of their time on the templated layer and 20% on the last mile. The market is about to invert which one pays.
The first question is this: Are you doing the templated middle of my work or the last mile of it? The next question is this: What’s the smallest step you can take this week to begin automating the templated middle and move toward the last mile?
Make no mistake, if you don’t, someone else will.
The Window
Every technology transition produces two populations. The people who develop the scarce capability while it’s still scarce, and the people who wait until it’s default and unrewarded.
Data scientist was an exotic title in 2012. By 2020, it was a baseline expectation for analytical roles. The people who learned it in 2013 caught a once-in-a-career wave. The people who waited until 2021 found themselves competing for shrinking margins against fresh graduates with the same skills.
Last-mile capability is on that curve right now. The forward-deployed posture, the embedded specialist, the contextual operator won’t stay rare. They’ll diffuse. The people who develop the capability now will catch the wave.
The people who lean in now will shape how this plays out.
