Are You Still the Master of Your Craft?

This is the third in a series exploring what the research actually says about AI and your career. Previously: AI displacement is real and which roles are most exposed. This week: what happens inside your role when AI arrives.

How AI is quietly rewriting what your role actually consists of

In the last two weeks we confirmed that AI-driven displacement is real and that which jobs are more prone depends on the makeup of your daily work tasks and how well-bounded your daily work is in terms of digital workflows where every step can be digitized, chained, and executed end-to-end.

This week, I want to look at what happens inside your role when AI arrives. Here's something nobody warned us about as AI crept into our workplace.

Let's recall how we've seen AI enter our work since ChatGPT launched in late 2022. It didn't announce itself. It showed up one task at a time. First it drafted emails you used to write from scratch. Then it handled the research synthesis that took half your afternoon. Then it started compiling reports, summarizing meetings, generating first versions of documents. Quietly, task by task, the craft you spent years building started being rewritten underneath you.

For most knowledge workers, this is already happening. The question that matters isn't whether AI is changing your work. It's which parts is it taking? The parts that make your craft valuable, or the parts that get in the way of it?

---

AI enters through your most describable tasks

The tasks AI takes first share a specific profile. They have clear inputs, defined outputs, and a process that can be described as a set of instructions. This isn't a guess. It follows directly from how large language models work: they respond to instructions, so the tasks they handle best are the ones that can be instructed (Eloundou et al., 2024).

Anthropic's Economic Index (January 2026) confirms this empirically. Their research team analyzed millions of anonymized conversations with Claude (Anthropic's AI assistant) and mapped each conversation to specific occupational tasks from the US Department of Labor's O*NET database, covering 19,265 work tasks across 923 occupations. The pattern in the data is striking: AI usage is concentrated in instruction-describable tasks across every knowledge work domain.

Some examples of what AI is being used for, heavily:

• Software: Fixing bugs, generating code from specifications, debugging errors
• Marketing: Drafting ad copy, compiling campaign reports, writing briefs
• Legal: Reviewing contracts against standard clauses, drafting correspondence
• Finance: Running scenario analyses from structured data, building models
• Administration: Classifying and categorizing emails, managing scheduling, processing invoices

And what AI is not being used for:

• Physical tasks requiring hands-on work or site presence
• Real-time interpersonal interactions (counseling, negotiation, de-escalation)
• Tasks requiring sensorimotor coordination or manual dexterity
• Crisis response and novel judgment under genuine ambiguity
• Relationship-dependent work (managing stakeholders, reading organizational dynamics)

The top 10 most common AI tasks account for 24% of all consumer Claude usage. The single most common task (fixing software bugs) represents 6% alone. On the enterprise side, the concentration is even sharper: the top 10 tasks represent 32% of all API traffic, with back-office automation (email processing, invoice handling, calendar management) growing fastest.

---

Which tasks AI takes determines whether your craft deepens or erodes

Most people hear "upskilling" and think: learning something new, a course on prompt engineering, a certification. But in the context of how AI is editing your role, upskilling and deskilling mean something more specific and more consequential.

Think of a chef. If AI takes over recipe creation, flavor pairing, and menu design (the creative, high-skill work) and leaves you chopping vegetables and plating to spec, you still work in the kitchen. But the craft that made you a chef is gone. That's deskilling: your role got simpler, even though your title didn't change and your paycheck looks the same.

Now reverse it. AI takes over inventory tracking, order management, and prep calculations. What's left for you is flavor development, menu innovation, and guest interaction. The routine cleared out. The craft stayed. That's upskilling: your role became more focused on the work that actually requires your expertise.

Anthropic's researchers measured this directly across the economy. They built a model predicting the education level required for each of the 19,265 tasks in O*NET. Then they removed the tasks AI is currently performing and asked: does the average skill level of what remains go up or down?

Here's what this looks like in practice:

Deskilling: AI takes the craft, leaves the routine

Upskilling: AI takes the routine, leaves the craft

The pattern is consistent: whether AI deskills or upskills your role depends on which of your tasks it takes. And since AI currently enters through the instruction-describable tasks first, the outcome depends on whether those happen to be your role's highest-skill work or its lowest-skill work.

And here's the part nobody talks about: the danger of surviving. Most displacement conversation focuses on the people who lose their jobs. But deskilling is arguably harder to recover from than a clean layoff. A layoff forces action. Deskilling lets you drift. You're still employed. Your paycheck still arrives. But your craft is eroding, your market value is declining, and by the time you notice, the gap between what you can do and what the market needs has widened quietly.

---

This is happening faster than any prior technology shift

You might be thinking: okay, but how much time do I have? The answer is less than you'd expect.

The World Economic Forum surveyed over 1,000 employers representing 14 million workers globally (Future of Jobs Report, January 2025). Today, employers estimate 47% of work tasks are performed mainly by humans alone, 30% through human-machine collaboration, and 22% mainly by technology. By 2030, they project roughly equal thirds: 33% human, 33% human+AI, 34% AI-only. That's a 14-percentage-point drop in human-only work in five years.

Honest context: employers have historically overestimated how fast they'll automate. In the 2020 WEF report, they predicted 47% automation by 2025. The actual number by 2023 had barely moved. So these projections are directional intent, not guaranteed timelines.

But the direction is consistent. 86% of employers expect AI to transform their business by 2030. 73% are accelerating task automation. And the pace of AI adoption itself is unprecedented. Anthropic's data shows that AI usage across US states is converging toward parity roughly 10x faster than previous economically significant technologies (Anthropic Economic Index, 2026). Shifts that historically took 50 years to diffuse are compressing into 2-5 years.

One important nuance: the human-in-the-loop dramatically extends what AI can handle. When people work with AI collaboratively (asking follow-ups, correcting course, breaking problems into steps), AI maintains effectiveness on tasks that would take a human many hours alone. In autonomous mode (enterprise systems where AI handles tasks end-to-end), effectiveness drops sharply beyond simpler, shorter tasks (Anthropic Economic Index, 2026). This is why the "human+AI" category in the WEF framework matters so much: it's where AI extends your capabilities rather than replacing them. The professionals who learn to collaborate with AI effectively will have a longer runway than those who either ignore it or fully delegate to it.

---

How to tell which edit you're getting

You don't need a time audit to figure this out. You just need to notice something that's probably already happened.

“Think about the last time AI saved you real time at work. Was it on the hardest part of your job, or the most tedious part?”

If AI handled the hard stuff (complex analysis, research synthesis, strategic drafting), it's doing your highest-skill work. What's left is the routine. That's the deskilling edit.

If AI handled the tedious stuff (formatting, data compilation, first drafts of boilerplate), it cleared the low-skill work off your plate. What's left is the judgment, the relationships, the decisions. That's the upskilling edit.

To go one level deeper, try these three questions. They take 60 seconds:

1. What did AI most recently save you time on at work? This tells you which tasks are already being taken.

2. What's the part of your job you could never hand to AI, even if it were perfect? This tells you what your durable core is.

3. If AI took over everything it currently can in your role, would the remaining work be more interesting or less interesting than what you do today? This gives you the verdict in your own words.

If your answers came easily and you felt clear about them, you probably know where you stand. If they felt murky, or if the honest answer to question three was "less interesting," that murkiness is worth paying attention to. It might mean your craft is being edited in a direction you haven't fully reckoned with.

---

Getting the full picture

The reflection above gives you directional clarity. For the rigorous, data-grounded version, I built Alignment Resilience. It runs this analysis against 19,265 human-rated work tasks, your actual resume, and live market data. You get a Resilience Score showing where you stand relative to your role's average, a task-by-task breakdown of what's exposed and what's durable, your strongest assets that AI can't replicate (with market demand trends), and 2-3 concrete career paths forward grounded in real hiring data.

Whether you use that clarity to reposition yourself internally (volunteering for projects that lean on your durable strengths, shifting your task mix toward the work AI can't touch) or externally (leading your job search with your AI-resilient assets instead of your most automatable skills), the goal is the same: make sure the edit AI is making to your role is one that deepens your craft, not one that erodes it.

Scoring is free. Upload your resume, answer 5 personalized questions, and see your score in about 15 minutes. No credit card required. If you want the full report with task breakdown, durable assets, career paths, and a personalized 30-day action plan, it's $79 (founding beta).

---

Research referenced:

1. Eloundou, Manning, Mishkin & Rock, "GPTs are GPTs: Labor Market Impact Potential of LLMs," Science, 2024
2. Anthropic, "The Anthropic Economic Index: Economic Primitives," January 2026
3. World Economic Forum, "Future of Jobs Report 2025," January 2025