Skills guide

Every other page on this site reads a number straight from Anthropic's published usage data. This page doesn't — it's general commentary on what tends to hold up alongside AI-heavy work, offered as context once you know where your own role sits.

Automation-heavy roles

When AI is doing more of the task on its own

Occupations where usage leans automation-style tend to involve tasks an AI can carry through largely unsupervised — routine drafting, standardized classification, data entry, and similar repeatable work. The skills that tend to matter most here aren't about out-competing the automation, but about sitting around it.

• Exception handling — knowing what to do when the automated output is wrong, incomplete, or doesn't fit the situation in front of you.
• Quality review and spot-checking — catching errors before they reach a customer, patient, or filing.
• Systems and process troubleshooting — understanding how the pieces connect well enough to fix something when it breaks.
• Translating output for people — explaining what an automated result means to a client, patient, or colleague who wasn't part of producing it.

Augmentation-heavy roles

When you stay in the loop with AI

Occupations where usage leans augmentation-style tend to involve iteration, feedback, and validation — the person keeps steering the work rather than handing it off. Here the AI behaves more like a tool you operate than a replacement for the task itself.

• Framing the problem well — the quality of AI-assisted work tracks closely with how clearly the task and context were specified going in.
• Critical evaluation of AI-generated drafts — reading output skeptically rather than accepting it at face value.
• Deep domain expertise — the knowledge that lets you notice when something plausible-sounding is actually wrong.
• Synthesis — combining several AI outputs or sources into one coherent result, which is often the part that isn't automated at all.

Low measured exposure today

A low score isn't a permanent guarantee

A large share of occupations in the dataset show almost no measured AI usage yet — often because the work is hands-on, in-person, or doesn't translate well into a text-based conversation with a model. That's a snapshot of current usage, not a ceiling: tooling has historically reached physical and in-person work eventually too, just on a different timeline and through different interfaces than a chat-based dataset would pick up.

• Watch adjacent, more-exposed occupations — automation often moves through a job family before it reaches every role in it.
• General digital fluency — comfort with new tools tends to transfer, even across a large gap in how "AI-touched" a role currently looks.
• Hard-to-automate-for-real-reasons work — physical dexterity, in-person trust, and situational judgment tend to hold their value longest, not because no one has tried, but because they're genuinely harder to replicate.

Across the board

Skills that tend to hold up regardless of pattern

A handful of skills show up in almost every version of this conversation, independent of whether a role leans automation or augmentation.

• Verification and critical thinking — treating any AI output, including this page's framing, as a claim to check rather than a fact to accept.
• Clear written communication — increasingly the actual interface between a person and an AI system, and between people reviewing AI-assisted work.
• Learning velocity — how quickly you can pick up a new tool or workflow tends to matter more than mastery of any single current tool.
• Baseline comfort with AI tools, even where your own occupation's exposure score is currently low.

Check where your own role actually stands

This page is deliberately general and isn't about your specific job. For a specific, data-backed answer, search your job title or browse the ranked occupations.