MIT Just Studied 20,000 Work Tasks. Here's What It Says About Your Job.

MIT researchers just published the largest study of its kind: 40 AI models, tested on over 20,000 real work tasks drawn from the US economy. Not hypothetical tasks. Actual tasks that actual workers do, evaluated by practitioners in those fields.

The findings should change how you think about AI and your career. Not because they're scary. Because they're more specific than most people assume.

It's not an apocalypse. It's a rising tide.

There's been this narrative that AI will suddenly take over entire jobs overnight — a "crashing wave." The MIT data doesn't support that. Instead, AI capabilities rise smoothly and predictably across most tasks. The researchers call it a rising tide.

That sounds calming, but it means something specific: if you know which tasks are in your job, you can watch the tide come in. You don't need to guess. The pattern is visible in the data.

The tasks that are going first

The study found that routine tasks in HR, logistics, and finance face the most near-term disruption. But here's the nuance most coverage misses — it's not entire jobs that are automated. It's components.

A financial analyst doesn't get replaced. The spreadsheet reconciliation part of their day does. An HR professional doesn't get replaced. The resume screening part does.

This is the part most professionals get wrong about AI risk. They think in terms of occupations. AI thinks in terms of tasks.

The counterintuitive finding: wages can actually rise

MIT economist David Autor's earlier work, which this study builds on, found something that runs against intuition: when automation eliminates the lower-expertise parts of a job, wages for the remaining workers tend to increase.

Why? Because fewer people are doing the job, but they're spending all their time on the parts that actually need expertise. The bar gets higher. The value per hour goes up.

The people who lose out are the ones who were mostly doing the parts that got automated. The people who win are the ones who were already spending time on judgment, relationships, and context.

So what should you actually do?

Here's the framework I recommend, based on what the study actually shows:

Map your task inventory. Write down every discrete task in your role — not vague responsibilities, actual tasks. "Generate the weekly sales report." "Respond to client emails about billing." "Sit in on the product planning meeting." Be specific.

Rate each task on two dimensions. First, how much expertise does this task require? Second, is it mostly text-based (the study's scope is text tasks — that's what LLMs do)? High-expertise, text-heavy tasks are where you want to spend your time going forward.

Let AI handle the rest, visibly. If AI takes over three hours of low-expertise work every week and you use those three hours to do something that requires judgment, you've just made yourself more valuable. Not despite AI — because of it.

Watch the tide. The MIT researchers found that AI capability gains are predictable. You can see them coming. That means you have time. The question isn't whether automation will reach your tasks. It's whether you'll reposition before it does.

The uncomfortable truth

The most at-risk professionals aren't in any particular industry. They're the ones who can't distinguish between the parts of their job that require real expertise and the parts that are just process. Process is automated. Expertise compounds.

Take the afternoon to map your tasks. It's the most career-defensive action you'll ever do that takes less than two hours.