Python typing survey: 'helps AI code' doubled to 38%, yet 30% aren't sure it does

AI coding tools have moved into Python typing fast. In the 2026 Python Typing Survey, published by Engineering at Meta on October 27, 38.4% of developers picked "makes AI-generated code more accurate" as a reason for type hints, roughly double last year's 18.5%. Yet when asked directly whether types help AI-assisted coding, 29.8% said they were not sure.

That gap is the survey's most interesting tension. Developers increasingly use types for and with AI, but have little evidence about whether it pays off.

AI writes types, runs checkers and fixes errors

The survey drew 1,794 responses between August 5 and September 2, 2026. Among the 1,517 people who answered that question, 42.5% had AI tools write type hints for them, and 61.0% used AI for at least one typing task, such as writing or explaining types.

About a quarter, 24.4%, said an AI coding tool runs their type checker for them. In that group, 61.6% also let AI fix the resulting errors, compared with 19.3% of everyone else.

Not all of it is welcome. Some respondents complained that AI writes types that are too broad, and one described agents that "spam #type: ignore comments everywhere" to make errors disappear.

No other benefit drew as much doubt. Runtime guarantees came second among the six listed, with just 7.7% unsure, about a quarter of the AI figure.

Help has moved from Stack Overflow to chatbots

Between 2024 and 2026, the share of respondents who turned to Stack Overflow for typing help fell from 46.4% to 22.9%. LLM chat climbed to 41.7%, up from 26.0%. Official Python typing documentation also gained, from 56.2% to 65.9%, as did checker-specific documentation, which reached 60.5%.

Usage is flat, but checkers are splitting

Frequent use of type hints barely moved across three years: 85.5% this year, against 85.9% in 2025 and 88.0% in 2024. Which checker people use changed much more. Half of those who named their checkers, 49.9%, now run two or more, up from 46.0% a year earlier, and speed was the top reason for choosing one, at 56.5%.

Running several checkers has a cost. Among multi-checker users who answered the question, 42.7% reported that their checkers disagreed with each other. The biggest complaint overall remained missing type information in third-party libraries, chosen by 65.1%, with Django, SQLAlchemy, Pydantic, NumPy, pandas and Polars named repeatedly.

The survey has limits worth keeping in mind. Respondents were self-selected through developer channels, and Meta, which ran it, also builds the Pyrefly checker, a point the write-up itself notes. Of 264 Pyrefly users, 83.3% also used another checker.