Your AI is a superb reasoner. It's also, left to its own devices, a confident guesser — because it can only work with what it already knows. The moment a good decision depends on what other people think, want, or remember, reasoning isn't enough. It has to ask.
That's the whole idea behind Widgy. Your AI designs a form, hands you a live link, and reads the answers straight back over MCP — no dashboard detour, no CSV export, no pasting rows into the chat. The loop is short enough to matter:
create_form, you share the link, people answer, and your AI calls list_form_responses to read structured results — then acts on them.The real unlock isn't building a form quickly — it's that your AI decides what belongs on it. Hand it a task ("help me price this job," "work out why people are churning") and it can reason backwards from the goal: to answer this well, what would I need to know that I don't? Those gaps become the questions. You never write the form; you describe the problem.
From there it builds the form and hands you a link — and you send it down whatever channel reaches the person: email, a text, a DM, a line in Slack. The answers land back in the chat, and now your AI is grounding its response in real data instead of filling the gaps with assumptions. A form stops being something you design and becomes a channel your AI opens whenever a task needs outside input.
"I'm not sure, let me guess" becomes "I'm not sure, let me ask." Here are three everyday jobs where that swap changes the answer your AI gives you.
Ask your AI what to work on next and it will happily reason from thin air. Ask a form first, and it reasons from your actual customers. "Should the next release be faster search or a mobile app?" stops being a coin toss the moment fifty people have ranked it.
A rating scale gives your AI numbers it can average; a long-text question gives it the sentences people actually wrote. When it reads the responses back, it isn't just counting votes — it's quoting the person who said "I'd pay double if it synced to my phone" and weighing that against the eleven people who shrugged at search. That's a prioritisation call made on evidence, with the receipts attached.
Here's a real one, not a hypothetical. This is the exact feedback form I'm running to build Widgy — seven questions, about two minutes, and (naturally) built in Widgy itself:
Notice the mix: multiple-choice to segment, ratings to score, long-text where the gold is, an email to follow up. One control type per job — your AI chooses each one.
When answers land, my AI doesn't hand me a spreadsheet. It averages the two rating questions, buckets the "what's missing" replies into themes, quotes the sharpest line verbatim, and pulls the emails of people who want in. "Eleven of fifteen rate their current tool a 3 or lower, and the same word — clunky — turns up in six answers." That's a build decision with evidence under it, not a hunch.
This isn't a mock-up — it's live, and I read every answer. Tell me what your form tool gets wrong. Question five literally asks you to be brutal.
Every good proposal starts with the boring questions: budget, timeline, what you're actually trying to do. Drag that out of someone over email and you get three half-answers across five messages. Put it in a form and your AI gets clean fields — a number for budget, a date for the deadline, a dropdown for company size — before it writes a single line.
With that structure in hand, the AI can qualify (is this a fit?), route (who should see it?), and tailor (draft the reply that speaks to their constraints). The difference between a generic "thanks for reaching out" and a proposal that already knows the budget is just a form the AI filled in for you.
Six fields, each a clean data point — so the reply your AI drafts already knows the budget, the deadline, and the ask before it writes a word.
Some decisions aren't yours alone — they're the average of what a group wants. When can everyone meet? Which of these three names do we ship? What are we each blocked on this week? These are miserable to gather by hand and easy to gather with a form.
Send the same short form to ten people and your AI reads back ten structured answers at once. Instead of you squinting at a thread, it does the tallying — "Thursday works for eight of ten, and the two nos both prefer Friday afternoon" — and hands you a recommendation you can act on, not a spreadsheet you have to interpret.
Ten teammates, ten tidy responses — your AI tallies the day, spots the theme in the priorities, and flags every blocker by name.
The same pattern stretches further than three examples. Audience research and surveys are the natural next step — point a form at a mailing list or a community, and let your AI roll up the ratings and pull the quotes that sharpen your positioning, pricing, or next post. Anywhere a decision would be better with a few real answers behind it, a form is the shortest path to getting them.
None of this asks you to build a form. That's the point — your AI builds it, shares it, and reads it back. You just decide what's worth asking.
Or, before you go — answer the form that's shaping what I build next.