Skip to main content
A/KAI Kills What?

Task // Software & data

High exposure · 1 occupations · 0 evidence

Writing new code

Implementing features and scripts from a specification.

Task automation exposure

70/100

High exposure

Exposure basis

[v4/median] score 70 (raw 70 | A 70 / B 85 / C 35 | modality: other) Modern code‑generation AIs (e.g., Copilot, GPT‑4) can translate specifications into functional scripts with high accuracy, but human oversight is still typically needed for debugging and integration. (capable: GitHub Copilot / OpenAI GPT‑4)

2 capabilities · 1 occupations

Basis

How exposure was assessed

[v4/median] score 70 (raw 70 | A 70 / B 85 / C 35 | modality: other) Modern code‑generation AIs (e.g., Copilot, GPT‑4) can translate specifications into functional scripts with high accuracy, but human oversight is still typically needed for debugging and integration. (capable: GitHub Copilot / OpenAI GPT‑4)

Automation exposure70/100

How much of this task's output current AI can produce or substantially assist.

Strongest applicable capability88/100

Best current model strength on any capability this task requires.

This is an assessment of technical feasibility plus published evidence - not a forecast. Even fully automatable tasks are often not automated, because of cost, verification, liability, regulation or the fact that the task is bundled with other work.

Capabilities

AI capabilities involved

  • 33 models tracked · strongest: GPT-6 Astra

  • 35 models tracked · strongest: GPT-6.1 Terra

Occupations

Occupations that perform this task (1)

Importance is the O*NET-style weight of this task within each occupation. The same task can be highly exposed but peripheral to one job and core to another.

Occupations performing Writing new code, with task importance and exposure.
OccupationIndustryImportanceRoleOccupation AI exposure
Research ScientistsScience & research3/5secondary49

Evidence

Evidence (0)

No evidence item is mapped to this specific task yet. The exposure figure above rests on the v1 editorial baseline and on the capability assessment, and should be treated as low-confidence. Pages with fewer than 0 evidence items are excluded from sitemaps.