# AI wrote the code. How hard do you review it?

Agents now write about half the code in a large 2026 survey. The open question is what humans still check before merging.

Poll by GloboPoll

## Results

With 8 votes as of 2026-10-20, the most voted option is “I read the whole diff, but not line by line” with 50%.

- I read every line before it merges — 0 (0.0 %)
- I read the whole diff, but not line by line — 4 (50.0 %)
- I skim the diff and lean on tests and CI — 2 (25.0 %)
- I rely on tests, CI and review bots, not on reading — 2 (25.0 %)
- If it runs and the task is done, it ships — 0 (0.0 %)
- I don’t let agents write code I merge — 0 (0.0 %)
- Other — 0 (0.0 %)

8 votes

## Poll facts

- Status: open, accepting votes
- Type: single choice
- Published on: 2026-10-13
- Closes: 2026-10-31
- Data updated on: 2026-10-20

## How votes are counted

One vote per person (per account, or per IP address and browser for anonymous votes). Anyone can take part: this is an open online poll, not a representative statistical sample. Results update live while the poll is open.

[Vote on GloboPoll](https://www.globopoll.com/en/polls/ai-wrote-the-code-how-hard-do-you-review-it)

---

Source: https://www.globopoll.com/en/polls/ai-wrote-the-code-how-hard-do-you-review-it