Task vs Job
Core distinction: The task is what you can describe in a brief: get the SKU, write the spec, build the XLS, render the email, ship the code. The job is the implicit knowledge that decides which task to do, in what order, with what content, for what reason. AI is rapidly absorbing tasks. Jobs are what’s left.
The articulation
Benedict Evans, May 2026, on enterprise AI: “What’s the task and what’s the job? What do you actually want? How do you split the music from the plastic?”
The task side is explicit, describable, often deterministic: “Get the SKU number. Build the XLS. Write the spec.” The job side is implicit, opinionated, often non-verbal: “What’s new and different? What do most people think sounds stupid? What’s the customer actually trying to do that they haven’t said?”
A great accountant in 1985 wasn’t great because of the arithmetic — the calculator did the arithmetic. They were great because of judgment: which ledgers matter, where the irregularity lives, what to flag. Spreadsheets ate the arithmetic. They didn’t eat the accountant.
Why this matters for AI strategy
Two questions follow:
-
What’s the task you do vs. what’s the job you do? If your customer is paying you for the task and the AI does it for free, you are exposed. If they’re paying you for the judgment, you’re not (yet).
-
In your own product, what’s the user’s task vs. their job? The thing the user clicks is the task. The decision underneath the click is the job. Most products only address tasks. AI lets you finally address jobs.
How to apply
Take any unit of work in your product or business and ask:
- The task layer: What does the work product look like? What are the inputs and outputs? What’s measurable, repeatable, describable?
- The job layer: What does the person actually want? What changes in the world if it works? What would they pay 10x for? What’s hard to articulate but obvious-in-retrospect?
The task layer is where most software has lived since 1980. The job layer is where humans have lived. AI is the first technology that can plausibly operate at the job layer (at varying levels of trust), which means the task layer is up for grabs in a way it wasn’t before.
Strategic implication
Sell the outcome of the job, not the completion of the task. The vendor selling “we’ll generate the campaign” is selling a task in a world where tasks are about to be free. The vendor selling “we’ll increase mobile-deposit adoption by X%” is selling a job. The first will commoditize. The second can’t, because nobody else can credibly underwrite the outcome.
This is also why “AI replaces software” gets it wrong. Software replaced tasks. AI is doing the same — but it doesn’t yet replace the job, because the job lives in implicit knowledge that nobody has written down. Build software that absorbs more of the job than your competitors do, and you win.
Failure modes
- Productizing the task while your customer pays for the job — you’ll keep shipping and they’ll keep buying, until a competitor productizes the job and you’re left selling commodity tasks.
- Mistaking output for outcome — measuring tasks completed instead of jobs done.
- Selling AI as a productivity tool — implies you’re shipping faster tasks, not better jobs. Productivity is a 1980 frame.
Cross-References
- Companion frames: Was-Cost-of-Task-Your-Moat, What-Was-Impossible-Now-Cheap, Absorb-Innovate-Disrupt
- Related patterns: Augmentation-Over-Automation, Build-AI-Run-Deterministic, Bottleneck-Shifts-Upstream