How do I author a study from my agent?

Once your agent is connected to Fiuto, you can ask it to draft a study from scratch in a single message. You describe what you want to learn; the agent drafts a plan and the blocks for it. The draft sits in your studies list until you decide it is ready to launch.

This article covers what to ask for, how the iteration usually shapes up, and the cleanup worth doing before you flip the switch to live. It applies to any connected agent: Claude Code, Cursor, Lovable, v0, Replit, or another MCP-ready tool.

If you haven’t connected an agent yet, see Connect your agent to Fiuto with our MCP first.

How to brief your agent

The shorter the brief, the more your agent fills in for you. The more specific the brief, the closer the first draft is to what you want. Three rough levels.

Quick brief

Say what you want to learn in a sentence.

“Create a quick five-question study about the new sign-up flow.”

The agent picks a sensible mix of blocks (a rating, an open-text, a couple of multiple-choice) and drafts the study. Good for casual or exploratory work.

Specific brief

Name the goal, the audience, and any blocks you want.

“Create a click test on this Figma prototype, then ask a rating question for clarity, then a short open-text for whoever rated below 4. Audience is designers.”

The agent follows the structure you described and uses the Figma URL you gave it.

Plan-first brief

Ask for the plan before the build.

“Plan a study to find out whether our pricing page is clear to first-time visitors. Do not build the blocks yet.”

The agent drafts a plan and saves it. You read it, edit it (in chat or in the web app’s Plan tab), and then ask for the build:

“Build the study from the plan.”

Plan-first is useful when the shape of the study is not yet clear, or when you want a teammate to review the plan before any blocks get drafted.

The typical iteration

Once a draft exists, you iterate in the same chat.

1. Ask for the draft. “Create a five-question study about the new onboarding flow.”
2. Read what the agent drafted. It lists the blocks back to you.
3. Adjust. “Drop the matrix question. Rewrite the first one, it is leading. Add a rating block for usefulness at the end.”
4. Preview. “Give me the preview link.” Open it in a browser and walk through the study as a respondent would.
5. Adjust again. Anything that felt off in the preview becomes the next round of edits.
6. Launch. “Launch the study and share the link.”

Each round of edits lands on the current state. You do not have to repeat the full brief.

Keeping the study in draft until ready

An unlaunched draft has no respondent link and cannot collect responses. It remains visible to people who already have access to the study in Fiuto.

Your agent will not launch a study on its own. If it thinks the draft is good, it might say so, but the launch happens only when you say it should. If you want to be extra cautious, you can also confirm at launch time:

“Are you sure this is ready? List anything you would change.”

Use this when the draft matters (going to a client, going wide to a public audience) and a second pass before launch is worth a minute.

Cleanup worth doing before launch

A short checklist that catches the most common day-of-launch regrets.

1. Read each question end to end. Easy when there are five questions; do it anyway. Wording reads differently to a respondent than to you.
2. Check any placeholder URLs or images. If the study uses a Figma prototype, an image, or any external asset, make sure the link or file is the right one and that it is reachable. Placeholders left in by mistake (a REPLACE_ME in a link, a stock image) block the study from going live.
3. Make sure the intake fits the audience. If you only want designers, the intake should screen out non-designers. If you want everyone, do not add a screener that filters them.
4. Preview the study. Click through as a respondent would. Anything that surprises you here will surprise them.
5. Check the total length. If the block count is climbing, ask the agent to trim anything that does not help answer the research question.
6. Sanity-check the visibility rules. If you added rules (“only show this follow-up to people who rated low”), test that the right respondents see the right blocks.

Your agent can help with any of these:

“Read each question and flag anything that sounds leading.” “Is the intake set up to screen out non-designers?” “Walk me through who sees what, given the visibility rules I added.”

Once the draft passes the checklist, ask your agent to launch.

When to switch to the web app

Your agent covers most of what you would do in the editor, but a few things are faster in the web app:

1. Tweaking the plan by hand. Open the Plan tab and edit the markdown directly. If the plan is locked, unlock it before editing. You can also ask your agent to revise the plan in chat.
2. Visual previews. The editor renders each block as a respondent would see it. The preview link does the same but for the whole study.
3. Reading responses at scale. Once results come in, open the version under Launches to skim distributions and read open-text answers side by side.

You can switch back and forth. Your agent reads from the latest state of the study, so an edit you made in the web app shows up the next time you ask it something. The two paths compose.

What to do if the draft is way off

1. The framing is wrong. The agent drafted a click test when you wanted a survey, or vice versa. Tell it what shape you want instead: “Switch this to a survey-only study. No click tests.”
2. The audience is wrong. Add a screener or change the existing one: “Replace the intake. I want product managers, not designers.”
3. The wording is consistently off. Ask for a register change: “Rewrite every question to sound more casual.” Or “Make these read like questions a researcher would ask, not a marketer.”
4. You changed your mind. Leave the draft unlaunched and start a new study with the new brief.

The iteration loop is cheap. Use it.

Next steps

Pull study results into your agent Read and analyse your study's responses from your agent. What landed, what's missing, what to ask next.

Launching a study The launch flow, what changes when a study goes live, and the live gate.

Editing the study plan The Plan tab. The chat-driven way to settle a study's shape before building blocks.