Karpathy indicates that white-collar jobs have a high AI risk score of 6.7, meaning your manager could be the next one.

Andrej Karpathy just detonated a stun grenade.

This legendary AI researcher—former Chief AI Officer of OpenAI and a leading figure in the machine learning community—released a detailed risk analysis in March 2026 that would send chills down the spine of any board of directors. His conclusion was straightforward: White-collar jobs with annual salaries exceeding HK$1 million face a 6.7 out of 10 risk level related to AI automation. Low-paying jobs only score 3.4. The numbers are jarring. The metaphors are even more jarring: your manager, your strategy consultant, every knowledge worker around you—are on the chopping block. Not in 2050. Right now.

Nobody wants to see a risk rating

& nbsp ;

Karpathy 's analysis, breaking down automation risks by salary level, reveals a shockingly stark correlation. High-paying roles requiring expertise, judgment, and abstract thinking—those we've always considered "safe"—rank highest on the AI replacement risk list. Why ? Because these are precisely the tasks AI agents excel at: analyzing complex documents, developing strategies, managing workflows, and making data-driven decisions. The research results show that:

--- 6.7 Risk Level: Professionals (Managers, Consultants, Analysts, Project Leaders) with an annual salary of HKD 1 million–2 million
HKD 600,000–1,000,000 ) --- **3.4 Risk Level:** Low-paying positions (annual salary HKD 300,000–600,000) Automation thrives on scale and abstraction. A single AI agent can replace the workload of multiple junior knowledge workers. One agent drafting invoices, reading complex queries, checking inventory, and queuing approvals effectively replaces dozens of hours of human labor. But you can't automate the work of a cleaner with a single algorithm—at least not yet. The painful truth: You're constantly training your own replacement. Every email you send, every memo you write, every decision you record—this is data for the AI agent to learn. Once it learns? It won't need you anymore.

67% of companies already know

& nbsp ;

Karpathy 's analysis didn't come out of thin air. Organizations are now moving with lightning speed. Latest data shows that 67% of Fortune 500 companies have now deployed AI agents in production environments—compared to only 34% a year ago. This is no longer a trend. This is a turning point.

And these are not toy experiments. These are real, live-fire production agents doing real work:

--- Drafting financial documents and proposals
--- Read and categorize customer queries
--- Check inventory and flag supply chain issues
--- Queuing for routine business decisions
--- Coordinate scheduling across teams

This isn't theory. This is happening right now, in the companies you've heard of. The fastest-moving organizations have already seen amazing results: Organizations adopting open AI policies have reported operating cost reductions of 20–35% in just six months. In other words: either there are fewer people but the workload hasn't decreased, or the workload is the same, but there are far fewer people. Think about what this means for promotion cycles, headcount, and your job security.

Claude Opus 4.6 changed the game

& nbsp ;

The real turning point came at the end of 2025: Claude Opus 4.6 introduced "AI Teams," capable of working autonomously without human intervention. These are not chatbots that can talk. They are not fancy automated tasks. They are teams of AI workers that can collaborate autonomously.

The intelligent agent team can:

--- After obtaining the business objectives, break them down into tasks yourself .
--- Execute these tasks in parallel or sequentially
--- Able to collaborate with other intelligent agents without requiring permission
--- Self-adjustment when problems arise
--- Record everything for audit trails

This is the moment we've been building since the invention of the Transformer. AI is no longer an assistant. It is the executor. Companies that adopt agent teams early on will see significant cuts in their white-collar salaries. Not all at once—that would cause labor unrest—but steadily. Quietly. One role after another. The jobs that will disappear first aren't necessarily the most creative or strategic. They'll be mid-level jobs: coordinators, mid-level managers, and those "process managers" who spend all day moving information between emails. Precisely because they provide structured and valuable services. But structure is what AI agents can most easily replicate.

The managerial crisis has arrived

& nbsp ;

Karpathy 's headline is carefully chosen but meaningful: he specifically points out that management faces the highest risk. Managers face a risk score of 7.1—higher than consultants. Why? Because management is essentially information processing:

--- Reading progress updated
--- Comprehensive report data
--- Schedule a meeting
--- Drafting the decision
- team coordination
--- Documentation results

Each of these tasks is something an AI agent can do. And do it better—without self-esteem issues, office politics, or Friday afternoon fatigue.

AI managers never have bad days. They are impartial. They don't miss background information because they're stuck in two-hour stand-up meetings. They handle everything, forget anything, and only promote what truly matters.

So if you're a manager reading this: congratulations. Your role is the most easily automated in the entire enterprise. If an AI agent can do five of your jobs in a quarter of the time, your company doesn't need three layers of you. The difficult part? You're assessed as replaceable even before you realize it.

Structural shift in hiring and staffing

& nbsp ;

This is what's happening now in forward-thinking organizations:

--- They haven't started layoffs yet. That would cause lawsuits, a talent drain crisis, and a media backlash .
--- They've simply stopped replacing departing staff. Someone's retiring? Don't fill the vacancies with manpower. Deploy AI agents instead; the cost is only one- tenth . Operations Manager," "AI Agent Team Leader," "AI-Human Workflow Expert. " These are names that sound important because traditional organizational structures are crumbling. --- They are preparing for a talent transformation. They've quietly launched a "skills enhancement" program, defining positions that require "AI collaboration" as a core skill. This is how you compress payrolls without issuing a restructuring announcement. Slow enough, the market won't react. Fast enough, Wall Street will notice the improved profit margins. What if you're not in a position where you're being actively "transformed" into a hybrid role? What if your job is still defined in the old way? Then you're in a 6.7 risk bucket, the clock ticking away.

What Karpathy didn't say (but should have been said)

& nbsp ;

Karpathy 's analysis focuses on risk scoring. But the real story is about leverage. AI doesn't replace workers. It amplifies the workers who remain. Who are the people who will prosper in the next 24–36 months?

--- Learn to lead AI agents, not humans
--- **Shift from execution to direction** (telling the agent what to do, rather than doing it itself )
--- **Developing Judgment Skills That AI Cannot Replicate:** Deciding what's important, what questions to ask, and when to break the routine . --- **Build expertise that is amplified by AI, rather than replaced by AI** In other words: The humans who remain won't be doing their old jobs. They'll be doing new jobs, with AI teammates.

But this transition is brutal for those who are unprepared. Most people are unprepared. Most people hope it's just a hype cycle. Most people are waiting to see what happens. Most people are in a 6.7-point risk bucket, sinking deeper every week.

Ironically: Replacement or Enhancement?

& nbsp ;

Here's a twist, something Karpathy hints at but most people miss: **risk isn't really replaced. It becomes redundant.** There is a difference. "Being replaced" means your job is taken by AI. "Becoming redundant" means your position no longer exists because smaller human teams are using AI as a lever to do those jobs. No drama. No lawsuits. Organizational structure is simply reconfigured around those who remain. Those who remain will be the early adopters. They will start working with AI agents now. They will learn to think like, "How can the AI agent do this task?" rather than "How do I do this task?" In other words: the safest role of the future is not your job today. It's a hybrid role created through learning and collaborating with AI. The companies that first figure this out—shifting from "AI replacing humans" to "AI amplifying the right people"—will dominate the next decade. Companies that resist or move slowly will also cut salaries, but painfully, through natural attrition and restructuring.

What should you do now

& nbsp ;

Karpathy 's analysis is sobering. But it's not a death sentence. It's a map. You know where the risks are. Now you must act.

If you are a manager: Stop viewing AI as a tool for your team. Start viewing it as a teammate. What happens if you delegate a complex project to an AI agent instead of a junior employee? What will you learn? What will break down? Since the agent frees up your time, what can you delegate upwards?

If you are an individual contributor: Your value is no longer your ability to execute tasks. It's your ability to ask the right questions, assess trade-offs, and direct work. Start practicing. Start asking your manager for opportunities to strategize, not just execute.

If you are a leader or founder: 67% This number should shock you, but not because competitors are ahead. It should motivate you because your window is closing—the window to thoughtfully integrate AI to strengthen your organization. In another 18 months, you'll be forced to compress your payroll under pressure rather than intentionally designing it.

Common theme? Stop preparing for replacements. Start preparing for enhancements.

Are you ready to prepare your team for the future?

UD's AI Staff solution helps you safely and strategically deploy AI roles alongside human teams. Don't wait to be disrupted to be forced to make decisions. Start the conversation now. 28 years of experience serving over 50,000 companies.