JustujuAI, layer by layer.Open in the interactive map

Copilots

AI built into the tools people already work in, such as code editors, documents and email, suggesting and increasingly acting while the person stays in charge.

Layer 9 · Applications & Interfaces · Run time · reviewed 2026-10-30

Read first: The agent loop, AI features inside products.

D1 · StoryFor young children and total beginners

Think first: who should have the final say, you or the helper?

A copilot sits next to the pilot and helps, but the pilot is still in charge.

AI copilots suggest text, code or answers, and you decide what to keep.

The helper suggests, you decide

  1. Helper suggests: text, code or an answer
  2. You decide: keep it, change it or bin it
  3. Your work: you stay in charge

Like: A copilot helping, while the pilot stays in charge.

D2 · PictureFor curious beginners

Copilots bring model calls into existing tools: code editors, email, documents and search. They are the deepest form of AI built into a product.

Good designs keep people in control: suggestions appear inline, are easy to accept or reject, and show their sources.

Many copilots now have an agent mode that hands a whole task to a coding agent or other agent, and comes back with changes to review.

Behind the interface sit the lower layers: an agent or orchestrated calls, a model endpoint, and infrastructure.

Like: A copilot helping, while the pilot stays in charge.
Where it breaks: A human copilot is accountable and consistent. An AI copilot can be confidently wrong in different ways each time, so review matters more, not less.

D5 · FrontierFor experts

An accepted suggestion looks like a productivity gain, but speed, correctness and skill can move in different directions.

Mixed evidence. Developers finished a set task 55.8% faster with an assistant in one controlled study (Peng et al., 2023), while experienced developers on their own projects took 19% longer (METR, 2025).

Uneven gains. In customer support, an assistant raised issues resolved per hour by about 15% on average, mostly for less experienced workers (Brynjolfsson et al., 2023).

Insecure suggestions. About 40% of 1,689 programs completed by an early code assistant were vulnerable (Pearce et al., 2021).

Over-reliance. In a survey of 319 knowledge workers, more trust in AI went with less critical thinking (Lee et al., 2025).

Examples as of 2026-10-30

Sources

  1. The Impact of AI on Developer Productivity: Evidence from GitHub Copilot (2023)
  2. Generative AI at Work (Brynjolfsson, Li, Raymond) (2023-04-23)
  3. Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity (METR) (2025-07-12)
  4. Asleep at the Keyboard? Assessing the Security of GitHub Copilot's Code Contributions (2021-08-20)
  5. The Impact of Generative AI on Critical Thinking: a survey of knowledge workers (Lee et al., CHI 2025, Microsoft Research) (2025)

How it connects

More in Layer 9 · Applications & Interfaces