What's next

Better questions, not just faster collection

The next real leverage point isn't more features on the collection side - it's helping researchers ask better questions in the first place. The plan is to ground survey creation in established interviewing methodology - the Mom Test, Jobs to Be Done - so create_survey can push back on leading questions or hypothetical framing before a survey ever goes out, instead of only after the fact.

Turning raw answers into a real picture

Right now, get_responses hands back raw text, one response per row. The goal is to go further: synthesizing what a set of open-ended answers is actually saying - patterns across respondents, something closer to a working "digital persona" of the people you're hearing from, not just a transcript to read line by line.

Taking the security warning seriously

The FAQ is upfront that survey content and responses aren't scanned for injected instructions yet - that's a real gap, not just a disclaimer. Closing it is a priority before this is anything more than a beta: scanning what a researcher puts into a survey, and what a respondent sends back, for attempts to manipulate the other side's assistant.

Beyond open text

V1 is deliberately open-ended text only - it's the format that best fits an assistant-mediated conversation. Closed and multiple-choice questions, and letting a respondent revisit or continue an answer instead of submitting once and being done, are both on the list once the open-text core is solid.

Paying for it properly

Credits exist today so usage isn't unlimited, but there's no self-serve payment - access during the beta comes from a manually issued invite code. Real, self-serve payment is planned once the core experience has held up to more than a handful of people using it.

None of this is a promise with dates attached - it's the honest list of what's next if the idea keeps holding up. Feedback shapes the order more than this list does.