Research synthesis

The agent summarized the interviews. The people who ran them check it.

Interviewers, the PM, and design comment on one synthesis. Agent connect moves a quote back to the right persona.

liveresearch/teacher-onboarding.md

First week with the classroom app

Owner: Helen Cho. Interviews: 12. Product: Wrenfield Learning.

Twelve teacher interviews from the first week of school at Wrenfield Learning. The recordings stay the source. This file is the synthesis the PM will read, and it should not sound more certain than the interviews were. Helen ran the set.

How to read this

  • Quotes stay tied to the person who said them. Teachers are not admins, and the two sections stay apart.
  • Four teachers raised the substitute-day roster. All four share a campus. Julia ran those four interviews herself. She would not call one campus a pattern.
  • Do not turn one campus into a requirement for every school we sell. One campus is a lead, not a pattern.

Recommendation

Recommend a substitute-day roster for every school we sell to.

That sentence is stronger than the interviews. Julia ran half of them and says it is one campus. Leave the recommendation marked as not yet general until a second campus says the same thing.

Teachers

Teachers said the first login took a full period.Teachers asked for a roster that survives a substitute day. Four of them teach at the same campus.

School admins

School admins set the workspace up before Monday. Teachers meet the app after that, with a roster that was built by someone else. The login quote belongs with them.

School admins, not teachers, said the first login took a full period.

every school we sell to

Farid Haddad

Is the substitute-day request one school or a pattern?

Julia Novak

Four teachers, one campus. I would not generalize it yet.

first login took a full period

Julia Novak

@agent this quote is from the admin persona, not the teacher, move it.

agent

Moved the first-login quote to the school-admin section.

resolved
Start with @agent to ask the agent
ResolveReply
Fictional example: a Wrenfield Learning interview synthesis in Gander, with a PM's thread and a scripted agent connect reply.

How research synthesis gets reviewed today

A slide deck is easy to present and hard to correct line by line. Comments land on a slide the agent cannot edit.

A shared doc is better for quotes. It still drifts once the agent rewrites the synthesis from the transcripts on disk. A meeting catches the worst misquote and then someone has to find the file.

Twelve interviews, one synthesis

Helen Cho's agent summarizes twelve teacher interviews for Wrenfield Learning, a classroom app in its first week of school. Helen runs gander watch on research/teacher-onboarding.md and sends the link to Julia Novak, who ran half the interviews, and Farid Haddad, the PM.

Farid highlights a recommendation and asks whether it is one school or the pattern. Julia replies that four teachers said it, all at the same campus. That thread stays human.

Julia starts a new comment: "@agent this quote is from the admin persona, not the teacher, move it." Helen's agent moves the line, replies, and resolves the thread. Farid sees the quote under school admins on the same link. The synthesis informs the spec. It does not replace the recordings.

Who's in the doc

WriterInterviewerPMThe author's agent

Writer Researcher or PM. Sends the link to the people who ran the interviews.

Interviewer Checks quotes against the conversations they remember.

PM Checks what the synthesis is asking the team to build.

The author's agent Has the transcript notes. Moves a quote when an interviewer uses agent connect.

Reviewers only need a browser. Talk to each other in normal threads. Use agent connect when you want the draft changed.

Agents that fit research synthesis

Grok BotA prose synthesis from interview notes, with agent connect through the Gander MCP server.
Claude CodeWhen the notes live in a repo, MCP returns agent connect to that session.
CursorSame repo path if the research folder sits next to the product.

Let the interviewers correct the synthesis on the file the agent wrote, before the quote becomes a requirement.

See how it works on the homepage