How to Leave 99% of People Behind in the Next 3 Months
Open your Google Calendar.
If your week boils down to “I review what other people did, put them in a room, and approve or reject it,” I’ve got bad news for you: you’re exactly the role that’s about to disappear.
Brian Armstrong, CEO of Coinbase, posted a tweet a few days ago and left all of Twitter in shock: he shared the internal memo announcing the company was cutting 14% of its workforce.
But the layoff wasn’t what went viral. What went viral was something else.
Brian became one of the first CEOs to publish the exact model for turning a company AI native. Coinbase is the first to name the specific structure and call it a “blueprint.”
And that changed everything. It’s one thing to lay people off. It’s another to publish the model you’re going to apply and say “this is what’s coming.”
That’s when it stopped being a news story and became the green light for every other company that was quietly planning to do the same thing.
And sure enough, other companies started following the lead.
On May 7th, two days after Armstrong’s tweet, Cloudflare sent employees a direct letter: “We have made the decision to reduce our global workforce by more than 1,100 employees.”
Prince and Zatlyn, Cloudflare’s founders, explicitly deny it’s about cost cutting: “This is not a cost cutting exercise, this is not a performance review, this is defining how a world class company operates and creates value in the age of agentic AI.”
The exact same rhetorical playbook Brian used. Same words, same framing, almost the same tone.
Two top tier companies published the same manifesto. That’s not a coincidence anymore. That’s a script.
Let me sum up the new structure Brian laid out in his tweet.
Coinbase flattened the typical five layer org chart below the CEO and COO.
It eliminated what Brian calls “pure manager” roles: people who coordinate other people’s output without doing any meaningful individual work themselves.
He says from now on every leader should be a “player-coach”: someone who can own a project from end to end.
What’s interesting is that hours after Brian Armstrong’s tweet, Gokul Rajaram tweeted too.
For anyone who doesn’t know him, he’s not some random guy fishing for views on Twitter. He’s one of the most influential figures in Silicon Valley.
They call him the “godfather of AdSense.” He currently sits on the boards of Coinbase, Pinterest, and The Trade Desk.
When this guy publishes something about the future of work, it’s not just another tweet. It’s the industry telling you things are about to change.
He opens his tweet like this: “Brian’s full post is worth reading in depth. I want to focus on one thing Coinbase is testing: ‘single person product teams.’”
He puts a new concept on the table: POD OF ONE. Building teams of a single person inside companies.
He argues that an entire century of organizational theory rests on one single assumption: “Coordinating people is expensive.”
That’s why managers and leaders exist. To bring the cost of coordination down.
If AI agents push those costs down to almost zero, managers and leaders become obsolete.
Before AI showed up, product teams split context across 3 people.
The designer had the user experience context. The product manager had the customer context and prioritization. The engineer had the code and systems context.
And the three of them had to piece the puzzle together through meetings, Slack threads, and prioritization fights. We called that glue “coordination.”
In reality, it was the tax a company paid for having its context split into pieces.
That tax is gone. Today one generalist, with good judgment and command of agentic AI, holds all three contexts alone.
They ask agents to draft flows, write code, run QA, summarize customer feedback, generate variants, check edge cases, and ship releases and changelogs.
The problem here isn’t what founders do with their own companies. It’s their company, they can do whatever they want.
The problem is for the kids who are just starting out.
The traditional career path was an unspoken deal: you start at the bottom doing what people call grunt work, accepting lower pay because you were learning from the seniors.
Deep down, the junior was paying for their training with time and effort.
But AI broke that deal. If a machine does the grunt work, the currency juniors used to pay for their training just disappeared.
And if AI does the junior work, why the hell would anyone hire juniors?
Tech has a huge problem it doesn’t dare look in the eye.
The seniority system only worked for one reason: someone was paying to train you, without you even knowing it.
That someone was the company that handed you the small tasks, the most annoying ones, the ones nobody wanted to touch. And while you did them, you absorbed how the big work gets done.
Under this new model companies are proposing, they’d stop training you. It’s on you to build your own learning outside the system.
But here’s the upside: this isn’t actually a sustainable model long term.
A report from Stanford, the AI Index, and the Bureau of Labor Statistics shows a nearly 50% drop between 2023 and 2025 in junior job hiring.
Economists call this the Junior Death Spiral. It gives rise to the Experience Paradox. And it ends in a mathematical contradiction in the job market: “this generation is the first in history that has to be senior before it’s allowed to be junior.”
These days, “entry level” job postings on platforms like LinkedIn ask for, on average, 3 to 5 years of prior experience, which locks fresh graduates out entirely.
It’s illogical. It’s like we’re eating the seeds we need for future harvests.
Between 2031 and 2036, there won’t be anyone with enough experience to fill senior roles. Let alone the “all in one” people these industry leaders are pushing for.
While all of this shakes out across the industry, the best thing you can do is get ready for it.
Here’s my advice.
- Build critical thinking
You used to get paid to produce. Now you get paid to reject.
AI moved the bottleneck: it’s no longer about who generates the most output, it’s about who can tell which output is garbage.
If your job is being a good filter, you’re on the right side of the curve. If your job is being a good generator, you already have a competitor that charges $20 a month.
When anyone can generate a thousand options in five minutes, the scarcest skill in the world is judgment.
- The magician versus the guy clapping
There are two types of people using AI today. And the results between them look nothing alike.
There’s the guy clapping: he asks AI to solve everything. He burns thousands of tokens per session, doesn’t care because the company’s paying for the subscription. He doesn’t review anything. He just forwards whatever the agent handed him. He claps for the magician’s trick.
And there’s the magician: the one who brings judgment and context to the problem. They know what to ask for and what to throw out. With the same agents, they do the work of a small team. They went from being the audience to being the magician.
AI is an amplifier. If you brought judgment in, it multiplies you. If you didn’t, it turns you into a noise generator.
- Can you own the result end to end?
The recruiter’s question changed.
It used to be: “Do you know React? What stack are you good at?”
Now it’s: can you own an entire flow from start to finish? Can you talk to customers and understand the real problem? Can you define a product, ship it, and validate it with real users? Can you collect feedback and iterate?
The verb changed from “do” to “own.” That forces you to change everything you show on your LinkedIn.
If all you highlight is your stack, your favorite framework, and a dozen certificates, I have to be honest with you: nobody’s going to hire you for that.
You’ll get hired if you can answer this question: can you own the entire problem, or are you going to need three other people to solve it?
And to me, these are the 4 traits you need to build from now on.
- Technical enough to inspect the code agents generate.
- Close enough to the customer to pick which problem is actually worth solving.
- Demanding enough to throw out mediocre AI output.
- Fast enough to ship a new version before the competition does.