Theme E: AI is now the default route in, through and around PyScript

What it is

Large language models as the medium of discovery, learning, construction and documentation for PyScript. This is not a single feature request; it is a shift in the environment, community and industry in which we operate.

What it means for PyScript

AI now sits on almost every path a user takes. Discovery: Nitau, an engineer, found PyScript because "the LLM told me that PyScript runs Python in the browser using WebAssembly." Building: Claudiu, a hobbyist, has moved from writing code to directing it, teaching Claude to use his PyScript helper functions and stepping back until "I'm never reading the code, because the code is being produced much faster than I can read it." Momin, an engineer, has built an entire platform around a copy-paste-from-LLM workflow for non-technical students. Sai, an informatician, runs an AI agent that generates Python executed in PyScript, and finds it competitive with a "billion-dollar company's" full coding-agent setup, partly because "PyScript on the browser is way faster, I can see what's happening."

There is a clear and important spread of attitudes, which we should represent faithfully rather than lose under the generic "AI" term. At one end, Claudiu is comfortable not reading the code. In the middle, Nitau uses "basic prompting" as "a conscious choice," remaining "the gatekeeper" who reviews everything before it enters his codebase, and Łukasz treats it as "a productivity booster, but sometimes it's more like a slot machine." At the other end, Kattni does not use AI at all, on ethical grounds (training data, climate, and what she called an "evangelical" culture), and also because her self-assessed Python knowledge means she "wouldn't be able to tell whether what I was just given is good or bad." Anna, a learner, deliberately avoids AI for schoolwork on principle while using it for lab work to move faster.

Two practical sub-findings deserve emphasis. First, model quality against PyScript changed materially and recently: Łukasz reported that before December 2025, using AI with PyScript was "pretty dangerous," and that with Opus 4.5 it became "tractable," speculating it may have been trained on the PyScript docs. Momin noted LLMs "cannot detect the current version of PyScript" and sometimes add random imports that break the code. Second, and strategically most important: our documentation increasingly reaches humans only after passing through an LLM. Nitau observed that he fed the docs' Markdown files to an AI and "it did a perfect job." Nicholas's own reflection, echoed to several interviewees, is that the people who read our documentation "are not people, they're LLMs."

This extends beyond documentation. As AI-native tooling (coding agents such as Claude Code and GitHub Copilot) becomes the default way practitioners build, the question is no longer only whether a human can read our docs, but whether an AI agent can correctly represent our APIs and services when a practitioner prompts it. How well we express our work to these tools increasingly determines the quality of response a practitioner receives about PyScript, and about Anaconda's products more widely. Nicholas has written in depth about the challenges this poses.

Future steps

Treat "how our resources are consumed by LLMs" as a first-class engineering, education and documentation problem (work Nicholas has already begun). Ensure the docs, API surface and examples are structured so an LLM produces correct, version-aware PyScript; the version-detection failure Momin reported is a concrete target. Keep a genuinely AI-free path fully supported and first-class, both because some valued community members (Kattni) require it and because learners (Anna) deliberately choose it. Avoid taking a single position on AI; the community spans the full range and trust depends on us respecting and embracing that. Treat how our APIs and services are represented inside AI-native coding tools as an extension of the documentation problem: what a coding agent generates about PyScript is now part of our public interface.

Standing across archetypes

Universal, but polarised. Engineers and hobbyists are furthest into AI-assisted building; educators are the most cautious; learners are thoughtfully selective.

Challenges

A definitional confusion has dogged discussion of AI and PyScript inside Anaconda. "AI in the browser" can mean two quite different things.

1. The first is running LLMs inside the browser runtime itself, using experimental web APIs.
2. The second is what every AI-using practitioner in these interviews actually does: use LLMs as ordinary and complementary tools (cloud services or agents) that generate or assist with Python, which then runs in PyScript.

Internal advocacy at Anaconda has focused on the first sense of "AI in the browser" (running models inside the browser); all the practitioner evidence in this report concerns the second (LLMs as external tools generating PyScript code or consuming PyScript-based resources). Acting on this theme therefore requires realigning internal direction with the evidence, rather than advocating for something with no demonstrated use case, market signal or community demand. That realignment is beyond the PyScript team's sole authority. Furthermore, were in-browser models ever wanted, JavaScript would be the better-performing tool for the job.