Google Smart Paste: AI‑Powered Context‑Aware Adjustments for Pasted Code

Developers frequently copy and paste code, but many paste actions require immediate edits such as fixing syntax or renaming variables, which disrupts workflow. An analysis of Google’s code repositories revealed that about 25% of paste events are edited right after pasting.

To address this, Google introduced Smart Paste, an internal tool that predicts the next state of the code environment and uses generative AI to make context‑aware adjustments to pasted code. The system leverages large sequence models like DIDACT to enable features such as code review and automated fixes.

Training data was collected from Google’s monorepo and logs, focusing on post‑paste edits near the original paste location. Heuristic methods filtered the data, resulting in a dataset where 28% of pastes required no further edits, allowing the model to learn when no changes are needed.

The model was fine‑tuned from a pre‑trained coding model (DIDACT) and calibrated to achieve a 65% exact‑match precision. A confidence threshold determines whether a suggestion is shown; suggestions below the threshold are discarded automatically.

Usability studies with ~4 000 engineers showed that only 6.9% of paste events triggered Smart Paste, but the acceptance rate was 42.5%, saving significant developer effort. Adjustments such as a 1.5‑second visibility window increased acceptance by ~15%.

Various visual modes were explored, including greyed‑out text, diff views, auto‑apply, and inline diffs. User feedback favored the inline diff approach, which balances control and visibility without overwhelming users.

Future work includes revisiting auto‑apply for high‑confidence small changes and extending the model’s suggestions beyond paste interactions, such as automatically updating all references after a rename.

In conclusion, Smart Paste provides AI‑generated, context‑aware code adjustments that are now enabled by default in Google’s most used IDEs, delivering thousands of suggestions weekly and markedly improving developer productivity.