Automation index

How much of a Prompt engineer’s job can AI already do?

A prompt and an eval come earlier. Knowing when the model lies, and what must not be automated, do not.

This percentage does not predict how many people will lose their jobs.

Task comparison

IS WORK OVER?

34%

AI can already do 34% of this job.

#21 / 763

Prompt engineer

Task weights are our estimates based on O*NET frequency data, not time shares published by O*NET.

The time saving for each task is estimated separately.

Task	Share	AI	Points
Design prompts, tools and chains. Estimate No public source provides a precise figure. This is our estimate.	24%	58%	+13.9
Evals, datasets and regressions. Estimate No public source provides a precise figure. This is our estimate.	18%	52%	+9.4
Integrate with product and safety limits. Estimate No public source provides a precise figure. This is our estimate.	16%	22%	+3.5
Talk to the business about a good output. Estimate No public source provides a precise figure. This is our estimate.	16%	18%	+2.9
Watch hallucinations, bias and cost. Estimate No public source provides a precise figure. This is our estimate.	14%	20%	+2.8
Decide what the model gets, and what it does not. Estimate No public source provides a precise figure. This is our estimate.	12%	16%	+1.9
Total	100%		34%

Modelled workweek: 100% · Public evidence: 0% · Estimate: 100% · Confidence: Low

IS WORK OVER?

22%

#378 / 763

Embedded developer

Task weights are our estimates based on O*NET frequency data, not time shares published by O*NET.

The time saving for each task is estimated separately.

Task	Share	AI	Points
Firmware, the peripheral and the buffer. Estimate No public source provides a precise figure. This is our estimate.	16%	40%	+6.4
Field, the hang and the log that is not there. Estimate No public source provides a precise figure. This is our estimate.	18%	14%	+2.5
Watchdog, the reset and not leaving it blind. Estimate No public source provides a precise figure. This is our estimate.	16%	28%	+4.5
Draw, the battery and not sleeping it wrong. Estimate No public source provides a precise figure. This is our estimate.	16%	22%	+3.5
Hardware, the revision and not fighting the pin. Estimate No public source provides a precise figure. This is our estimate.	18%	18%	+3.2
That it does not hang, and not leaving the watchdog blind. Estimate No public source provides a precise figure. This is our estimate.	16%	12%	+1.9
Total	100%		22%

Modelled workweek: 100% · Public evidence: 0% · Estimate: 100% · Confidence: Low