Python Typing Survey 2026: AI Use Rises as Developers Mix Checkers

Python type hint use barely moved across three annual surveys, but the tools around it are changing quickly. In the 2026 survey, developers reported using several checkers in the same workflow, relying more often on AI tools for typing tasks, and still running into missing or inaccurate type information in libraries.

The survey received 1,794 responses, most from experienced Python developers: 76.4% had used Python for at least five years. Its results offer a view of what developers need from type checkers, library maintainers and documentation authors. They also show why stable adoption rates don’t mean the typing ecosystem itself is standing still.

AI has entered the typing workflow

AI tools now help write annotations, explain type errors and produce changes in response to checker feedback. Of 1,517 respondents asked about AI and typing, 42.5% said AI tools had written type hints for them, while 32.6% used types to improve AI-generated code. Across four listed typing-related activities, 61.0% reported using AI for at least one.

The share who said typing makes AI-generated code more accurate or reliable rose from 18.5% in 2025 to 38.4% in 2026. That suggests developers increasingly see annotations and checking as part of reviewing generated code. Yet it doesn’t settle how much the tools improve results: 29.8% of 1,551 respondents were unsure how useful types are for better AI-assisted coding. That uncertainty was nearly four times the rate for runtime guarantees, the next least certain benefit.

Some developers described a tight loop in which an AI coding tool runs a checker and then tries to fix its errors. Among the 1,517 respondents who said how they run a checker, 24.4% said an AI coding tool invoked it. Of that group of 370 respondents, 61.6% also used AI to fix checker errors, compared with 19.3% of other respondents.

That workflow can save manual effort, but respondents also flagged familiar failure modes: generated annotations that are too broad and agents that silence errors with #type: ignore. One respondent said they correct AI-written hints because the tool “uses object for everything.” The Pyrefly team has begun studying how checkers and AI coding tools interact, presenting Type Checking in Agentic Workflows at the 2026 Python Typing Summit. More evidence will be needed to show which combinations help developers catch errors without adding noise.

Library type gaps remain the leading complaint

Missing type information in third-party libraries was the most commonly reported difficulty, selected by 65.1% of 1,486 respondents. Another 44.6% said the type system couldn’t describe everything their code does. These are related but distinct problems: one calls for better annotations in dependencies, while the other points to limits in expressing dynamic behavior.

Even libraries with dedicated type information can be difficult to check accurately. Of 156 people who wrote in their own answer, 22 mentioned dynamic constructs, decorators, metaprogramming or framework behavior. Ten described challenges involving arrays, dataframes or scientific typing. At least eight cited annotations they considered inaccurate, incomplete or too broad. Django, SQLAlchemy, Pydantic, NumPy, pandas and Polars came up repeatedly.

As one respondent put it, “Django does so much black magic it’s a wonder a type checker works at all.” Frameworks often rely on runtime behavior that’s hard to describe with existing typing features, so adding annotations alone may not address every mismatch. The survey points to two possible avenues: extend the type system where it falls short, and improve checker support for patterns used by particular libraries. The Pyrefly team has been working with Quansight on third-party library typing, including contributions to pandas-stubs and NumPy.

How often someone uses hints also changes which problems they notice. Among people who used type hints rarely or sometimes, 41.6% found the type system too complex or verbose; among frequent users, the figure was 14.5%. Missing library types showed the reverse pattern: 36.2% for less frequent users and 70.0% for those who used hints often or always.

Experience with typing itself didn’t map neatly onto complexity complaints. The share reporting complexity was 18.2% for respondents with zero to two years of typing experience, 16.0% for those with three to six years, and 21.0% for those with seven or more. Many newer typing users weren’t new to Python: 87.1% of the 396 people in the zero-to-two-year group had used Python for at least three years, and 77.3% used hints often or always.

That pattern has practical implications for learning materials. Developers may need guidance on introducing types into an existing codebase, rather than tutorials that assume they’re learning Python from scratch. Occasional users need a first checker run that’s easy to configure and errors they can understand. The survey’s most requested missing feature, better Callable syntax, was selected by 45.7% of respondents who answered that question. It ranked highly among people who reported complexity and among those who didn’t, making it a plausible focus for standards work—provided common cases stay straightforward.

Checker choice is shifting faster than typing habits

Frequent type hint use remained in a narrow range: 88.0% in 2024, 85.9% in 2025 and 85.5% in 2026. Each survey asked a different group, so these figures don’t track the same developers over time. Still, they suggest that broad adoption has been relatively steady.

Checker choices changed more. Speed or performance was the leading reason for choosing a checker, selected by 56.5%. Among respondents choosing from the same list of checkers in both 2025 and 2026, 49.9% selected at least two in 2026, up from 46.0% the year before. The 2024 survey had already captured frustration with checker speed and gaps in compatibility with Mypy plugins, alongside differences in checker behavior.

Multiple checkers can serve different roles. Of 700 respondents who selected Mypy, 297 also selected its editor plugin, while another 314 selected a different named checker integration without selecting that plugin. Most people who selected Pyrefly also chose another checker: 220 of 264, or 83.3%.

Written responses describe teams splitting checks between local editors and continuous integration. One team used different tools to suit members’ IDEs, then relied on Mypy for linting and CI. Another used Pyright for niche patterns in local development and ran Pyrefly in CI, turning to Pyright on files where Pyrefly failed as a way to improve checking performance.

Running multiple tools can introduce its own friction. Of all 1,486 respondents asked about difficulties, 33.3% reported inconsistencies between checkers. Among the 738 respondents who selected multiple static checkers and answered that question, the figure was 42.7%. Clear documentation of differences between checkers could help teams decide which disagreements need attention and which reflect deliberate variation.

Editor feedback and documentation shape the experience

Readability and documentation were the most commonly selected reasons to use type hints, at 93.6% of 1,571 respondents. Editor support also scored highly: 84.8% of 1,494 respondents who rated it gave it a 4 or 5. Among 103 people who supplied their own reason, about 39% mentioned completion, navigation, hover information or editor error messages. And 82.9% of 1,517 respondents ran a checker in an editor; VS Code was the most selected environment, at 59.3%.

The places developers seek help are shifting too. From 2024 to 2026, Stack Overflow use fell from 46.4% to 22.9%, while large language model chat rose from 26.0% to 41.7%. Python typing documentation grew from 56.2% to 65.9%, and checker-specific documentation reached 60.5%, up from 52.2% after a dip in 2025. General Python documentation edged down from 66.7% to 62.2%.

As more developers turn to typing and checker documentation, concise explanations matter. A useful checker error should point toward a short description and an example that addresses the specific issue. Those same references can help AI chat tools draw on relevant material when explaining a diagnostic. The full results and data are available in the Python Typing Survey 2026 download.

The survey ran from August 5 through September 2, 2026, with support from the Pyrefly team at Meta, the PyCharm team at JetBrains, the Python team at Microsoft, the ty team at Astral/OpenAI, and community contributors on discuss.python.org. It was shared through developer communities, newsletters, forums and checker project channels. The next useful measure of progress will be whether library annotations and checker feedback become easier to trust in real codebases, especially those that depend on dynamic frameworks or combine several tools.

The post Python Typing Survey 2026: AI Tools and the Shifting Checker Landscape appeared first on Engineering at Meta.