Task vs. Job Automation Test
Before predicting a role's automation, ask whether the automatable task IS the job or just one piece of it
- Difficulty
- Moderate
- Time to result
- ~days to results
- Steps
- 4
- Confidence
- 90%
A decision test for judging whether a new technology will eliminate a job or merely change it. Instead of scoring a profession by 'what percentage of its tasks can be automated,' you ask a sharper question: is the hard part of the job the visible task (writing the code, making the deck) or is it everything around that task? Where the task is the whole job it gets automated away; where the task is just the deliverable, automating it changes the job without removing it.
Origin
Articulated by Benedict Evans in his 'AI is eating the world' presentation, building on the classic economics distinction between tasks and occupations and his critique of the US government's O*NET job-exposure scoring.
Core principles
- 01A job is not a stack of separable tasks you can score line by line
- 02Sometimes the task IS the job (the elevator button); usually it is not
- 03What you actually pay a consultant, lawyer, or engineer for is rarely the visible artifact they hand you
- 04The deliverable (the deck, the code) is the task; the judgment about what to build is the job
How to run it
- 1
Name the visible task
Identify the concrete, automatable output of the role — the PowerPoint, the lines of code, the spreadsheet, the button press.
Pro tip If a demo can already produce a crappy version of this artifact, you have correctly found the task.
- 2
Ask what people are actually hiring for
Separate the artifact from the reason someone pays. You hire Bain not for a 75-slide deck but to walk your enterprise, understand the politics, talk to your customers, and work out what you actually need to do.
Watch out Do not confuse a cheap AI replica of the artifact with the value; 'that's not what you paid them for.'
- 3
Decide: is the task the job, or not?
If the task was the entire job (elevator attendant pressing a lever), the role disappears. If the task was one input to a larger act of judgment, the role changes and often grows.
- 4
Reject percentage-exposure scoring
Refuse the temptation to say '17% of a senior partner's work can be automated.' Professions cannot be decomposed that cleanly — it is the same failure as expert-systems trying to build a cat detector from 700 hand-coded rules.
Watch out Percentage-of-work-automated numbers feel rigorous but are, in Evans's words, deluded — treat them as noise.
In the wild
Evans lives in a building with a manually-operated elevator run by an attendant pulling a lever. After the 1950s these were automated into a button — a case where the task genuinely was the whole job, so the job vanished.
→ Correctly classified as a task-is-the-job role that automation eliminated outright.
AI grifters post 'I made a McKinsey deck with Claude,' but the deck was never the product. What Bain is paid for is to walk the whole company, understand its politics, and talk to real customers.
→ The consultant's job survives because the artifact was only the task, not the job.
When spreadsheets automated a week of manual calculation into 30 seconds, the intuitive prediction was fewer accountants. Instead the number of accountants rose through the 20th and into the 21st century.
→ Automation of the task expanded rather than eliminated the profession.
Common mistakes
Scoring professions by percentage automatable
Trying to break a role into tasks and label each automatable/not reproduces the expert-systems failure and produces confident but meaningless numbers like '17% of a law partner's work.'
Mistaking a cheap replica of the artifact for the value
AI can produce a bad version of the deck or the code, but the artifact was never why the client paid — you have automated the task while leaving the job untouched.
Is it for you?
Best for
Founders, analysts, and professionals assessing which roles or industries AI will actually displace versus merely reshape
Not ideal for
Situations where the automatable task genuinely is the entire job, in which case simpler displacement logic applies
From the transcript
“what's the hard part of the job? Is the hard part of the job writing the code line by line?”
“Is it the task or the job?”
“sometimes the task is the job. Like the classic example is like a an elevator attendant.”
“You can't kind of look at a senior partner at a law firm and say, well, 17% of their work could be automated like this…”
“What you actually pay Bane to do is to go and walk all over your enterprise”
From the episode
A rational conversation on where AI is actually going
Benedict Evans