April 2026 · 4 min read
There are two ways to ship AI in a research tool. You can treat the researcher as an obstacle to automate around — or as the expert whose judgment the AI exists to serve. We picked the second. Here's what that means in practice.
Principle 1 — Suggestion, never override
Every AI action in Tabular Pro is a suggestion the researcher confirms, edits, or rejects. The platform never silently mutates your data, re-codes a variable, or removes a response. Auto-magic is convenient right until it's wrong — and in research, wrong is expensive.
Principle 2 — Explainability is non-negotiable
When the AI flags a respondent as a speeder, you see the completion time, the median for comparable questions, and the rule that triggered the flag. When it proposes a codeframe, you see sample verbatims for each category. No black boxes. If we can't explain why, we don't ship it.
Principle 3 — Learning stays project-scoped
The AI learns from your corrections within a project — your codebook, your definitions, your edge cases. Nothing leaks between clients. Nothing trains a global model on your study. Your data is yours.
Principle 4 — Speed where it's safe, caution where it matters
Drafting a questionnaire? The AI moves fast — it's easy to edit a draft. Removing a respondent from the dataset? The AI proposes, you decide. We calibrate the autonomy of each feature to the reversibility of the action.
Principle 5 — Honest about limits
The AI is genuinely good at some things (open-end coding, quality flagging, draft reports) and genuinely weak at others (causal interpretation, strategic recommendation). We tell you which is which. No oversold "insights" that are really just restated frequencies.
Research is a craft. Our job is to make the tools sharper, not to replace the craftsperson.