Python survey finds AI tools are writing more of the types meant to keep code honest

AI coding tools are becoming part of Python’s type-checking workflow: a new survey found that 42.5% of respondents used AI to write type hints, up from 18.5% who said types made AI-generated code more reliable in last year’s survey. The results offer a useful snapshot of developers trying to make generated code safer, while showing how much of the benefit remains an open question.

Watch Desk analysis

What happened

The 2026 Python Typing Survey received 1,794 responses. Among 1,517 respondents asked about AI and typing, 42.5% said AI tools wrote type hints, while 32.6% said they wrote types to improve AI-generated code. Overall, 61% reported using AI for at least one of the typing-related activities covered by the survey.

There is a more hands-on wrinkle: 24.4% of 1,517 respondents said an AI coding tool invoked their type checker. Among those 370 respondents, 61.6% also used AI to fix checker errors, compared with 19.3% of everyone else. The survey was published by Meta Engineering, with support from teams at Meta, JetBrains, Microsoft and Astral/OpenAI.

Why it matters

Type hints can give developers and their tools more information about what code is supposed to do. Pairing them with a checker that catches inconsistencies could help expose mistakes in generated code before they travel further. But this survey records what developers do and believe, not whether AI-written hints reliably improve software.

The results also show the plumbing is not friction-free. Missing type information in libraries was the most common difficulty, selected by 65.1% of 1,486 respondents. And among respondents who used multiple checkers, 42.7% reported inconsistencies between them. More hints are not automatically better if they are vague, wrong or interpreted differently by different tools.

Our read

The interesting shift is not simply that developers are asking AI to write more code. They are putting type checkers into the loop, then asking AI to respond to what those checkers find. That is a promising workflow, but the survey leaves the key question open: does it produce better code, or just faster-looking code with a tidier annotation layer? For now, treat the figures as a snapshot of practice, not a verdict on effectiveness.

What to watch

• Whether future studies test how AI-written types affect real code quality and reliability.
• Whether checkers get better at explaining errors in ways both developers and AI tools can use.
• Whether missing and inconsistent library types remain the main obstacle as AI-assisted workflows spread.

Discussion spark: Should AI coding tools be required to run a type checker and show how they handled its errors, or is that too much process for a workflow that developers should control?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.