# Write a Discovery Interview Brief Before You Wing the User Call

*By AgentOps Lab*

You already paste support threads, hallway comments, and sales notes into an agent. What still fails is starting the interview, the recruiting plan, or the “users want X” summary before anyone agrees what the conversation is *for*. A messy signal goes into a blank chat; a confident script comes out; and only then do you notice the learning goal was assumed, the questions were leading, and the next step was a cold list you should never send.

This post is a **discovery-interview-brief ritual** for PMs, founders, and product operators who already use AI agents on products they own or are authorized to research. It is not a “find PMF” guide, not an income promise, and not a license to survey-spam or cold-outreach people who did not consent.

You will leave with a brief shape, a 15-minute drill, and a pointer to other live AgentOps Lab brief rituals. The Product Discovery pack store listing is not up yet, so this post does not send you to a checkout page that is still unpublished.

---

## The symptom: the blank-chat tax

Typical week:

1. Monday — you dump a support thread into chat and ask for “interview questions.”
2. Tuesday — the agent invents who to call and a willingness-to-pay question nobody agreed to ask.
3. Wednesday — the learning goal is “validate the idea,” which is three interviews hiding in one sentence.
4. Thursday — a outreach draft appears that would message people off a scraped list.
5. Friday — you redo the brief because nobody wrote down the goal, the do-not-ask list, or a consent path.

That is not “the model being dumb.” It is **missing discovery-ops**. An interview needs a job description the agent can follow: restate the signal → name one learning goal → say who could answer it (from the text you actually have) → list in-scope and do-not-ask topics → sketch a session and a consent reminder → draft a few non-leading questions → flag authorization and PII.

Skip that ritual and every later step — synthesis, a JTBD statement, or a decision to pursue — is guessing.

---

## Brief shape you can paste into a skill

```markdown
Restated signal (1 sentence):
Reported facts vs assumptions (labeled):
Primary learning goal:
Who to interview (role / context / why):
Channel (owned or authorized only):
In-scope topics:
Out-of-scope / do-not-ask:
Session shape (length / format / consent reminder):
Interview questions (≤5, non-leading):
Ethics check (consent / PII / authorization):
Next action:
Unknowns:
```

**Rule:** the agent must fill **one learning goal**, **who (or UNKNOWN)**, and **do-not-ask** before it offers outreach copy or a survey blast. If those are empty, the brief is theater. If you have no consented or authorized path to talk to anyone, stop — do not invent a cold list.

---

## Five fields people skip (and why they matter)

**One learning goal.** “Learn about onboarding and pricing and the competitor” is three interviews. Park the extras. A usable goal sounds like a question this session could actually help answer.

**Who, from the signal only.** Role, context, and why they could answer — not a demographic cartoon and not a purchased list. If the paste does not say who, write UNKNOWN and gather authorized context. Reach people only on owned or authorized channels, with consent.

**Do-not-ask, written down.** Pricing pressure, private life, deanonymizing other people, and competitors’ private data do not belong in a first conversation unless you have a clear, consented reason. Writing the boundary is what keeps the agent from “just adding a few conversion questions.”

**Non-leading questions.** Prefer “tell me about the last time…” over “wouldn’t you want a dashboard that…?” Five questions is enough for one session. A survey blast is a different job, and this ritual is not that kit.

**Facts vs assumptions, plus consent.** A quoted complaint in the thread is a fact. “Users in this segment will pay” is an assumption and should be labeled. If the paste includes names, emails, or secrets, redact them before the brief is reused. You own outreach, recording policy, and consent — the agent should not invent a recording rule you never stated.

A next action sits beside these: schedule-with-consent, refine-the-goal, gather-authorized-context, park, or stop. One signal, one next action. The agent drafts; you own the invitation.

---

## 15-minute drill (use a sanitized signal)

1. Pick one signal from a product you own or are authorized to research — a support thread, a user message, a sales note, or an observation. Strip names, emails, tokens, and secrets.
2. If you have no consented or authorized path to talk to anyone, stop this drill. Do not draft cold outreach.
3. Paste the brief shape (or a `discovery-interview-brief` skill file, if you have one loaded) and the sanitized signal into your agent.
4. Require one learning goal, who-or-UNKNOWN, and a do-not-ask list before any question set or email draft.
5. Accept ≤5 non-leading questions and one next action. If the agent invents participants, writes “users said” lines that were not in the paste, or proposes a survey blast, reject that output.

Success looks like a one-screen brief you could hand to another researcher — not a message the agent sent for you.

---

## Hard rules worth keeping in the skill

- Only research people who consented, or users of products you own or are authorized to research
- Owned or authorized channels only; not a survey-spam or cold-outreach kit
- Do not invent interview quotes, participants, or evidence
- Do not deanonymize anyone; sanitize PII and secrets
- No PMF, fundraising, revenue, or “you will ship a hit” promises
- You advise; the researcher owns outreach, consent, and what they ship

These rules are the difference between a reusable brief and a confident, unusable script.

---

## What this ritual is not

It will not find product-market fit, raise a round, or help you ship a hit. Files and rituals do not promise that.

It is not a license to research people who did not consent, scrape private data, or work on products you have no right to study.

It is not a substitute for judgment. A clear brief still needs you to decide whom to invite, how to obtain consent, and when to park the work.

---

## Where to go next

The same brief-before-blank-chat idea shows up in other AgentOps Lab rituals that *are* already listed. If you review pull requests more often than you interview users, the live Coding Agent Ops teaser walks a PR-review brief (intent, blast radius, risks, test gaps) — that is a merge-safety ritual, not a product-discovery interview pack:

→ [Coding Agent Ops free teaser on itch](https://moneyprinter-ops.itch.io/coding-agent-ops-free-teaser)

Catalog and other live packs:

→ [AgentOps Lab](https://agentops-lab.netlify.app/)

A full Product Discovery Ops pack (interview briefs plus synthesis, JTBD statements, opportunity briefs, assumption maps, lite experiments, and weekly decision logs) is being prepared. This article does not link a store or product page that is not live yet. When that listing exists, it will appear on the AgentOps Lab hub.

No PMF, fundraising, or earnings promises. Consented or authorized research only. Not a survey-spam kit. Do not invent quotes. Files only; you remain the researcher.

---

*AgentOps Lab builds installable ops skills for AI agents. Deep links only — we do not ask you to trust anonymous store roots.*