Do types make your AI write better code? 30% of developers have no idea.
Meta's 2026 Python Typing Survey: "When asked how useful types are for better AI-assisted coding, 29.8% of 1,551 respondents chose 'I'm not sure.'" Meanwhile "24.4% said an AI coding tool invoked their checker", and one developer complained of "Lazy coding agents that spam #type: ignore comments everywhere." Stack Overflow use fell from 46.4% to 22.9% as LLM chat rose from 26.0% to 41.7%, and Mypy fell from 59.5% to 46.1%. Meta's Pyrefly team tested it at the PyCon US 2026 Typing Summit: with the type checker run after every edit, their agent's success on well-typed code went from 79.6% to 83.9% with 21% fewer steps, and there was no measurable help on lightly typed code.
Source: "Python Typing Survey 2026: AI Tools and the Shifting Checker Landscape", Engineering at Meta, Oct 27, 2026 (Chris Wiltz) https://engineering.fb.com/2026/10/27/developer-tools/python-typing-survey-2026-ai-tools-checker-landscape/ Test numbers: "Type Checking in Agentic Workflows", Conner Nilsen, PyCon US 2026 Typing Summit, via Bernát Gábor's recap: https://bernat.tech/posts/pycon-us-2026-typing-summit-recap/
The editor, its code, the agent's caret and the Stack Overflow page are dramatizations; every quote and number on screen is from the sources. Pyrefly is developed by Meta, which runs the survey.
Has your agent ever written type: ignore to make the checker go green?
Transcript
Do types make your AI write better code? 30% of developers have no idea. Made up? It's in Meta's own article on its Python typing survey. Nearly four times the doubt of any other benefit. It matters because the agent's already in the loop. A quarter let AI run their checker. Most let it fix the errors, too. But one developer wrote, "Lazy coding agents that spam type: ignore comments everywhere. Green, without fixing a single thing." So, who checks the AI? Stack Overflow use halved. AI chat took its place, and even mypy fell from 60 to 46%. But Meta tested it. With the checker run after every edit, their agents solved more tasks in a fifth fewer steps. But only on well-typed code. On loosely typed code, no measurable help. Every type: ignore switches off what makes your agent better. So, next time it goes green, ask…
