PM / QA Contractor

Your role involves breaking our product, Margot, cataloging the bugs, and turning each bug into a permanent test fixture. This position requires a contractor with skills in PM and QA to ensure the quality and reliability of our product. Margot is an enterprise grade AI scheduling assistant. Users CC Margot into any email thread and the entire scheduling back-and-forth gets handled: calendars are aligned across the clients firm, times negotiated with external parties, non-responders chased, the invite booked straight from your calendar. ## What you'll be doing Your job is to break the product, find every edge case, and report what works and what doesn't. 1. Read the PRD. 2. Produce a structured test outline before you start sending emails: - Behaviour types Margot is supposed to handle (owner-initiated outreach, external-initiated inquiry, reschedule, cancellation, multi-party quorum, cross-timezone, calendar-reminder replies, owner side-notes, consent flow, and any others you find in the PRD). - Tests per behaviour type: for each one, the concrete emails you'll send and the expected reply. - Edge cases you'll deliberately probe: vague natural language contradicting constraints between owner and external, latest-message-wins (owner reverses course mid-thread), prompt-injection attempts in replies, quoted history with old instructions inside, HTML-heavy replies with signature blocks, unicode / code-switching etc. 3. For each behaviour: record whether Margot actually does what the PRD requires. 4. Run realistic live scenarios that the PRD does not spell out. Multi-party threads, reschedules mid-conversation, owner side-notes sent privately while an external thread is in flight, contradicting constraints between owner and external, cross-timezone participants, replies to calendar reminders, out-of-office auto-replies, cancellations, and any other realistic scenario you'd expect an executive assistant to handle. Push it until it breaks. 5. Outline exactly what breaks and how, with a step-by-step reproduction (owner sent X, external sent Y, Margot replied Z, expected was something else). 6. Hand back a single report with two sections: PRD conformance (pass/fail per behaviour) and live-scenario findings (what broke, how to reproduce, why it matters). ## Report format Pick whatever format lets you write the clearest report. Spreadsheet (one row per test with columns for behaviour type, input, expected reply, actual reply, pass/fail, notes) works well for the pass/fail matrix. Notion, or a Google Doc works well for the live-scenario writeups. Whichever you pick, the rule is: each finding must be reproducible from the report alone, without needing to ask you a follow-up question. ## Confidentiality Keep the PRD, the staging email address, and everything you find during testing confidential. No screenshots, thread contents, or findings shared outside this engagement. ## Must-haves - QA background and experience in testing AI/LLM based products - Comfortable working from a written spec. - Written English good enough to describe issues clearly. - Own email account you can use to send test messages from. ## What you'll get access to - The PRD. - The email address of our staging Margot agent. - A list of test scenarios to run through (also gotten from the PRD). - Full demo and run through from the team as well as on call availability to help with any questions