AI SkillsMay 20, 2026·4 min read

Amodei Says AI Will Handle 90% of Your Job. Here's How to Figure Out What's Left.

Anthropic's CEO flipped his forecast: the 10% AI can't automate doesn't shrink — it expands to fill your whole role. The catch is that most people haven't identified their 10% yet. Here's how.

By Forge Team

The uncomfortable thing about Dario Amodei's latest forecast isn't the 90% — it's the 10%. Most people are focused on the wrong number.

What Amodei said — and why the reframe matters

In May, Anthropic's CEO publicly shifted position. A year earlier he'd warned that AI could eliminate half of entry-level white-collar work. Now he's invoking the Jevons Paradox: "If you automate 90% of the job, then everyone does the 10% — and the 10% kind of expands to be 100% of what people do." (Amodei, public remarks, May 5)

He immediately added a caveat that most summaries skipped: AI is moving faster than the historical transitions this logic is drawn from, which means the window for rebalancing may be shorter than it was for, say, the spreadsheet replacing bookkeepers.

The Jevons Paradox is a real economic pattern. When steam engines became more efficient, coal consumption went up — not down — because lower costs created more demand. When AI automates 90% of a task, demand for the remaining human judgment tends to increase. But only for people who've developed it.

The skill implication

The 10% is not "soft skills." That framing is used to mean everything, so it means nothing. The 10% is specific:

  • Deciding whether AI's output is good enough to act on — or needs another pass before it reaches a real person
  • Translating between what a stakeholder wants and what AI can actually execute
  • Knowing when a task should stay human, regardless of whether AI could technically do it
  • Catching errors before they compound across an automated chain

These aren't personality traits. They're learnable skills. And most people aren't practicing them deliberately — because most of their time is still going toward the 90%.

What this looks like for an HR business partner

Maria is an HR business partner at a 220-person SaaS company. Over the last year, AI has absorbed roughly half her previous workweek: writing job descriptions, summarizing candidate packets, drafting offer letters, pulling compliance summaries, prepping performance review templates.

She finds the remaining work heavier, not lighter. Every conversation she's in now matters more — because the parts that don't require her judgment have already been handled.

Her 10% looks like this: reading the subtext in a manager's complaint during a retention conversation. Deciding whether a performance issue is a coaching problem or a structural one the manager created. Noticing that a DEI metric is trending wrong in one team — and knowing that data point needs a human conversation before any policy response.

These aren't things she can outsource. But she's also not practicing them deliberately. She's doing them, which is different.

Decide whether to use AI for a specific task — or keep it human.

The counterpoint: "relationships" isn't an answer

The most common response when people are asked about their 10% is "relationships." It's not technically wrong. It's also not useful.

A supply chain coordinator at a 60-person import/export firm said her irreplaceable value was vendor relationships. When asked to describe the last time she made a judgment call in one of those relationships that AI couldn't have made — she paused.

The real answer was specific: when a shipment from a supplier in Guangzhou is delayed and the explanation is plausible but inconsistent with a pattern she's seen twice before — that recognition, combined with knowing exactly which escalation to make and when, is her 10%.

That's not a relationship. That's a specific evaluative skill she's built over years of exposure to anomalous situations. "Relationships" as an answer tells you nothing about whether you're developing the right capability or simply assuming it will stay relevant.

How to find your 10%

Three questions that actually clarify it:

In the last month, what decisions did you make that you couldn't explain to AI in a way that would get the same output? Those decisions are in your 10%.

What AI outputs do you check before sending — and which ones do you just forward? What you check reveals where you still trust your judgment more than AI's.

If AI handled 90% of your current tasks starting Monday, which 10% would your manager notice missing first? That's where your professional value is currently concentrated.

Map a recurring task and identify which part should stay with you.

Amodei thinks the 10% expands to fill your role — but only if you've figured out what your 10% is and are actively getting better at it.

Like this post?

Get the next one in your inbox. Practical AI skills, no filler.