How to Survive the Shift to AI: The Skills Your Team Needs Now (Or Get Left Behind)

The 25-minute meeting

A few weeks ago I was on a call with members of my team. We needed to do something simple. Generate an SSH key on each person’s laptop and paste the public key into a chat thread so I could add it to a server. Three minutes per person. Maybe four if someone had never done it.

Before we started, I asked who was comfortable in a terminal. Multiple people said they were good. “Yeah, I use it all the time.” Cool.

It took us twenty-five minutes.

Not because the command was hard. The command is one line. It took twenty-five minutes because people couldn’t figure out how to copy text out of a terminal window. Couldn’t use arrow keys to recall the last command. Pressed CTRL+C and got confused when nothing happened the way they expected. Tried to right-click. Tried to drag-select. Pasted the wrong half of the key. One person closed the terminal by accident and didn’t know how to reopen it.

These are smart people. They build campaigns. They run ad accounts. They write briefs that clients pay for. And they got stuck on copy-paste in a black rectangle.

I’m not telling this story to dunk on anyone. I’m telling it because that meeting is the most honest diagnostic I’ve gotten on the state of my own company in the last two years. The gap between “I’m good with computers” and “I can actually do work with computers in a way that compounds with AI” is enormous. And it’s the gap that decides who has a desk job in five years.

This post is what I told my team afterward, expanded.

Two things are true at once

This isn’t a complaint and it isn’t a lecture. It’s a diagnosis. Both of these statements are true at the same time, and you have to hold both of them in your head without flinching.

The first thing that’s true is that there are real, structural, generational reasons the gap exists. Most people on my team grew up on iPhones. Their first computer was a touchscreen. They’ve never been told by an operating system, “I don’t understand that command, try again.” They’ve spent their entire adult lives inside walled gardens that hide the file system, hide the network stack, and hide the actual machine. That’s not their fault. That’s a product decision Apple and Google made, and it worked, and it sold a lot of phones.

The second thing that’s true is that none of that matters anymore. The work changed. The tools changed. AI ate the boring middle of office work. And whether or not the gap exists for fair reasons, the gap still has to close, because the work on the other side of it is the only work that pays.

You can be sympathetic to why and still be honest about what.

The “digital native” myth, with receipts

For about a decade, every consultant deck assumed Gen Z would be the most technically capable workforce in history. They grew up with technology. Of course they’d be good at it.

Then computer science professors started writing about what they were actually seeing in classrooms. In 2021, The Verge published a piece that quoted multiple professors who’d realized their students didn’t understand file directories [3]. An astrophysics professor at Cal Poly Humboldt was running a simulation lab and her students kept getting “file not found” errors. She asked where they’d saved their work. They didn’t know. They didn’t understand the question. The concept of “a place where the file lives” wasn’t in their model of computers.

That piece set off years of follow-up. A George Mason astronomy professor said he now spends class time teaching students about file extensions and terminal commands before he can teach them astronomy [3]. The Stack Overflow podcast did a whole episode on it [4]. PC Gamer wrote about it. The Verge’s reporter pointed out that a generation raised on search bars and apps doesn’t see hard drives the way older users do, because operating systems hide the file structure on purpose.

This isn’t a story about Gen Z being lazy or dumb. They’re not. The astrophysics students are doing astrophysics. The point is that “grew up with technology” turned out to mean “grew up with the abstraction layer,” and the abstraction layer is exactly what AI tools require you to peel back.

To get real value out of Claude Code or Cursor or any agentic system, you need to understand:

• where files live
• how processes run
• what a path is
• what an environment variable is
• how to copy something out of one window and into another without losing your mind

These are 1995 skills. And a lot of people who can swipe through TikTok at 60 frames per second have never been forced to learn them.

Why millennials are weirdly well positioned

I’m a millennial. I grew up on DOS, Windows 3.1, 95, 98, ME, XP, and 7, in that order. Switched to Mac in 2006. Built and broke and rebuilt machines with parts from local shops. Ran Linux on a laptop in college because the Wi-Fi card cost less if you bought one without an OS. I’m at home in a terminal. I’m also at home in a Figma file.

I’m not bragging. I’m describing a specific accident of timing. Anyone born roughly between 1982 and 1995 had to learn the underlying systems because the abstraction layer didn’t exist yet, and then we got to enjoy the fancy abstraction layer when it arrived. We can drop down a level when we need to. We can read an error message that includes a stack trace and not panic. We can SSH into a box. We can also use Notion and Linear without a manual.

Gen X and older sometimes dismiss AI as hype because they remember three previous cycles of “this changes everything” that didn’t. Fair instinct, wrong call this time. Gen Z grew up on the abstraction layer and often can’t drop below it, which means they get stuck the moment AI tools demand a real working knowledge of computers.

Millennials are uniquely positioned to bridge those worlds. That advantage is worth approximately zero if we don’t actually use it.

What AI just ate

Goldman Sachs put a number on this in early 2023 and updated it through 2025 and 2026. About 300 million full-time jobs globally are exposed to automation by generative AI [1]. In the US specifically, the technology can automate tasks that account for roughly 25% of all work hours. Office and administrative support has the highest exposure at 46% of tasks. Legal work is 44%. Architecture and engineering is 37%. Business and financial operations is 35% [5].

Goldman’s economist Joseph Briggs estimates 6 to 7% of US workers will be displaced over the transition, with younger entry-level knowledge workers hit first [1]. An April 2026 Goldman note found AI had already cut monthly US payroll growth by about 16,000 jobs in the prior twelve months [6].

Anthropic’s own data, drawn from millions of real Claude conversations, makes it more concrete. Their Economic Index ranks occupations by how much of their tasks Claude is actually doing right now. Computer Programmers top the list at 75% task coverage. Customer Service Representatives are at 70%. Data Entry Keyers are at 67%. Translators, Copy Writers, Tax Preparers, Bookkeepers, all clustered in that same band [2].

If your team’s job description has significant overlap with that list, the work is changing under your feet whether anyone tells you or not.

I want to make this less abstract. A few weeks ago I got a long message from an offshore SEO contractor we’d been working with. The message was an unprompted explanation of why his deliverables hadn’t been performing. It walked through the playbook he was running. Manual citation building. Free guest post outreach with no responses. Some link insertion services. The kind of work that maybe moved the needle in 2017 and definitely doesn’t in 2026.

The message itself was clearly written by AI. You could tell from the rhythm.

I’m not telling you that story to embarrass one person. The contractor’s individual situation isn’t the point. The point is the entire category. Manual citation building, free guest post outreach, basic on-page edits, lightweight backlink work, pasting boilerplate into outreach templates. All of it is now done faster, cheaper, and with better US-business context by tools like Search Atlas, Surfer, and direct Claude prompts. A category of work that supported tens of thousands of contractors globally just got eaten by software that costs $99 a month.

The honest response, if you’re in that category, is to either retrain into something AI can’t do yet, or to specialize in the parts of SEO that genuinely require judgment (technical audits on edge-case CMSes, real PR outreach, content strategy tied to brand voice). Pretending nothing changed and continuing to ship 2017-era work is the worst option, and it’s also the most common one.

The MIT NANDA report from 2025 confirmed this pattern in the data. The lead author told Axios there weren’t widespread layoffs from AI yet. What was happening instead was that companies were quietly not renewing their offshore contracts. “Jobs most impacted were already low priority or outsourced” [7]. The first wave of AI displacement is global outsourcing, not US W2s. That’s a temporary reprieve, not a permanent one.

What AI hasn’t eaten yet

Here’s where I want to be clear-eyed instead of doom-y.

Plenty of work is still safe. The Anthropic data shows roughly 30% of workers had zero AI task coverage in their occupations, because the tasks happen too rarely in digital form to even be measured [2]. That list includes Cooks, Motorcycle Mechanics, Lifeguards, Bartenders, Dishwashers, Dressing Room Attendants. Goldman’s analysis put building maintenance at 1% exposure and installation/repair at 4%.

If you can fix a transmission, frame a wall, run wire, weld a rail, or run a kitchen line at 8pm on a Saturday, you are extremely fine. Probably the most fine you’ve been in twenty years.

Inside knowledge work, what survives is a specific list:

• Real client relationships, where someone trusts you because you’ve shown up for them
• Strategic synthesis across messy, contradictory inputs
• Judgment under genuine uncertainty
• Taste, in the design or copy or product sense
• In-person trust, the kind that closes a deal in a room
• Coordinating other humans, including being the person who decides what AI should do
• Truly creative work that requires lived experience or new ideas, not remixed ones
• Anything physical, anything regulated heavily, anything where the cost of a wrong answer is somebody’s life or somebody’s lawsuit

Notice that almost none of those are the day-to-day tasks of a typical agency role circa 2022. Almost all of them are higher-leverage, more senior, more relationship-driven, and require more, not fewer, hard skills underneath. The middle of the pyramid is the part that’s collapsing.

The new ceiling for one operator

I have a friend in the industry, a peer of mine who’s a couple of years older. We came up around the same time. He’s gone deep on AI in a way I’ve only partially matched, and he’s worth describing because he’s the most concrete picture I have of what one person can do now.

He runs five Claude Max accounts. Five. Two hundred dollars a month each. He uses them as parallel workers, kicking off long-running tasks across different projects so he never has to wait on a single conversation. He builds Telegram bots for clients so they can edit their own websites by sending a message. He generated 22 SVG videos using Remotion in two hours, total cost about 86 cents. He shipped two small iPhone apps over a single weekend last month. He’s running a real product company, plus a few side products with paying users (Tridesk, ProxyBox, VoiceTrail, SalesConnector).

His own line on it: “I spend $60K a month on humans. Spending $1K a month on Claude that works around the clock and never complains is a no-brainer.”

He’s not a special genius. The tools are special. He’s just been willing to live in them. Every weekend for the last 18 months he’s been building things, breaking things, learning the rough edges, and getting faster. The compounding is brutal at this point. The gap between him and someone who refuses to install Claude Code is no longer a skill gap. It’s a category gap.

That’s the new ceiling. Not for a 50-person dev shop. For one person.

If you’re running an agency, a consulting practice, or a small product team, that’s the bar your competitors are about to clear. The boutique two-person shop that knows how to operate this way will out-deliver a fifteen-person agency that doesn’t. We’ve already watched it happen in copywriting. It’s happening right now in design and front-end. It’s coming for SEO, for paid media operations, for finance back-office, for legal review.

The skills your team actually needs

This is the playbook part. It’s also the part most teams fail at, because the skills feel basic and unglamorous and people skip them. Don’t skip them.

These are the ten skills. Each one is the floor, not the ceiling. If someone on your team can’t do these, they can’t use AI tools well, full stop.