Command Center
Aria chat_bubble Talk to Aria
Briefing

Jaime ROVO Agents — Portfolio + Build Plan

Three-agent ROVO Studio portfolio for Jaime's UX research org. Specced from the discovery interview Jaime ran on 2026-04-29. Build order, hard gates, and open-question triage so Topher can clear blockers in priority sequence.

Updated April 29, 2026 (v1 — discovery synthesized; Intake spec ready, Synthesis gated on legal, Digest deferred)

📌

TL;DR — What This Is

Jaime is Lead UX Researcher supporting ~35 designers across multiple platforms in a distributed model. Designers run their own studies; Jaime advises on methodology but rarely sees synthesis before stakeholders do. Two pain points score 5/5: synthesis quality variance, and Jaime's lack of execution visibility.

Jaime ran the discovery interview on Claude Chat (Haiku) on 2026-04-29 against the prompt at memory/drawers/day-job/jaime-rovo-agent-discovery.md. The export surfaced three agent candidates. This page is Aria's spec synthesis: build order, knowledge-base needs, hard gates, and Topher's open questions routed into hard-gate / build-time / deferrable buckets.

Phase 1 — Intake Agent

Ready to build. No participant data, no compliance gates. Standardizes designer intake; produces a methodologically sound research plan + study design that Jaime reviews before UserZoom setup. Closes the "Jaime has no consistent gate" friction.

Phase 2 — Synthesis Agent

Spec ready, blocked. Hard gates: SyF legal sign-off that platform consent covers AI synthesis (Q12.6), and a decision on UserZoom/UserTesting input mechanics (Q12.1). Highest single-agent value (5/5 quality + 5/5 visibility), don't start Studio until cleared.

Phase 3 — Quarterly Digest

Deferred to Q4 2026 earliest. Aggregates Synthesis output into an org-wide research video newsletter. Building before Synthesis has produced ≥10 outputs + a queryable archive exists = a tool nobody uses.

🧩

The Portfolio (3 agents)

1. Intake Agent

Phase 1 · Ready

What it does: Designer brings a research need. Agent interviews them one question at a time — goal, audience, design artifact, constraints. Recommends methodology with NNG citations. Produces a complete research plan + study design (tasks, discussion guide, quant metrics, success criteria) for Jaime's review.

Pain it relieves: 3/5 ad hoc intake; 3/5 freestyle plans; 5/5 Jaime visibility
Compliance: Clean — no participant data
Primary user: Designer; Jaime reviews
Build prerequisite: ~1 day authoring methodology Confluence pages

Spec: memory/drawers/day-job/jaime-rovo-agents/intake-agent.md

2. Synthesis Agent

Phase 2 · Gated

What it does: Designer hands the agent UserZoom/UserTesting raw data (transcripts + metrics CSV). Agent extracts qualitative themes by frequency (≥3 of N=5-7 = finding; below = anecdotal). Calculates SUS/NPR/UEQ/single-ease with confidence intervals. Pulls anonymized quotes with video timestamps. Flags contradictions for Jaime's review. Output is phased across Scenarios because the 8K cap is binding for full studies.

Pain it relieves: 5/5 quality variance; 5/5 Jaime visibility; 4/5 scattered findings
Compliance: Hard gate — legal must clear AI synthesis under platform consent
Primary user: Designer; Jaime reviews + edits
Build prerequisite: Legal yes + UserZoom/UserTesting input mechanics decided

Spec: memory/drawers/day-job/jaime-rovo-agents/synthesis-agent.md

3. Quarterly Digest Agent

Phase 3 · Deferred

What it does (when revisited): Aggregates all Synthesis Agent outputs from the past quarter, surfaces cross-platform themes, and produces a quarterly research video-newsletter outline for org-wide leadership.

Earliest revisit: Q4 2026
Build trigger: Synthesis live ≥1 quarter; ≥10 study outputs in queryable archive

Spec: memory/drawers/day-job/jaime-rovo-agents/quarterly-digest-agent.md

❓

Section 12 Open Questions — Triaged

🔴 Hard gates — block the build entirely

Q12.6 — Participant consent

Does our existing UserZoom (and incoming UserTesting) participant consent cover AI-driven synthesis by Atlassian ROVO? Affects: Synthesis Agent. Path: email SyF legal/compliance for written yes; ~1-2 weeks turnaround.

Q12.1 — UserZoom/UserTesting input mechanics

Native ROVO connector or manual export? Check the April 2026 connector list. Affects: Synthesis Agent input shape. If neither has a connector, designer manually exports + uploads transcripts + metrics CSV. Decide before authoring the agent identity prompt.

🟡 Build-time decisions — defaults below if Topher doesn't override

Question Default Override trigger
Q12.2 Video clips Output [HH:MM:SS] timestamps + URLs; designer pulls manually UserTesting exposes a clip-by-timestamp API
Q12.3 Figma deck output Markdown text; designer pastes into Figma Jaime asks for Figma-plugin JSON after v1
Q12.5 Knowledge base Confluence pages authored for retrieval (same pattern as Rooster) IT objects to write volume → Jaime's personal Confluence space
Q12.7 Model selection ROVO Studio default (April 2026 backend) Token-cap or video-handling limit hit
Q12.8 Token cap (Synthesis) Phased Scenarios: Quant → Qual → Anecdotal → Recommendations Atlassian raises 8K cap
Q12.9 Review interface ROVO Chat — designer @-mentions Jaime when output ready Workflow proves too loose → Jira-issue review pattern

⚪ Deferrable — don't block v1

  • •Q12.4 Confluence write-back / archive destination. All v1 outputs exit as text. Archive lands in UserTesting's repo when migration completes. Decide piping after migration.
  • •Q12.10 Quarterly Digest scoping. Whole agent is deferred to Q4 2026 anyway.
📚

Knowledge Base — Confluence Pages To Author

These pages need to exist before either agent is useful. Author in a Jaime-owned UX research space, marked for ROVO retrieval. Short headings, named relationships, no buried context (per the "human-readable but not AI-retrieval-ready" lesson from Rooster). Estimated authoring time: ~1 day with Jaime — she has the content in her head; the work is structuring it for retrieval.

Methodology library

  • • Card Sort
  • • Tree Test
  • • Click Test (First-Click)
  • • Task-Based Testing
  • • Moderated vs. Unmoderated
  • • Think-Aloud Protocol
  • • Sample Size Guidance

Quantitative methods

  • • SUS Scoring
  • • NPR Calculation
  • • UEQ Scoring
  • • Single-Ease Question
  • • Confidence Intervals + MOE
  • • Sample-Size Adequacy

Synthesis + standards

  • • Thematic Coding Standard
  • • Output Template (Jaime's format)
  • • Quote Anonymization
  • • Research Plan Template
  • • Findings Format Examples
🛠️

Build Sequence — Topher's Concrete Steps

Phase 1: Intake Agent (next 1-2 weeks)

  1. Topher + Jaime spend ~1 day authoring the methodology + quant + standards Confluence pages above
  2. Topher creates Intake Agent in ROVO Studio (Private visibility) using the spec at intake-agent.md
  3. Wire Confluence knowledge sources
  4. Test with 2-3 historical research-need scenarios Jaime supplies
  5. Iterate identity prompt + Scenario routing until Jaime signs off
  6. Switch to Team visibility; invite 2-3 pilot designers (Jaime picks)
  7. Pilot for 2 weeks; collect feedback in a Confluence retro page; iterate
  8. After retro: roll org-wide, or hold and refine

Phase 2: Synthesis Agent (only after legal + integration cleared)

  1. Topher emails SyF legal: "Does our UserZoom/UserTesting platform consent cover AI-driven synthesis by Atlassian ROVO?" — wait for written yes
  2. Topher checks ROVO connector list for UserZoom or UserTesting native connector. If absent, design assumes manual upload.
  3. Topher creates Synthesis Agent in ROVO Studio per spec at synthesis-agent.md
  4. Wire knowledge sources (already authored in Phase 1)
  5. Test with one sanitized historical study; verify quant math matches Jaime's prior manual calc
  6. Iterate; pilot with the same 2-3 designers piloting Intake
  7. Roll org-wide only after Jaime is confident she can mentor at scale on agent outputs

Phase 3: Quarterly Digest (Q4 2026 earliest)

Revisit when Phase 2 has been live ≥1 quarter and Synthesis Agent has produced ≥10 study outputs.

🛡️

Compliance Summary

  • ✓Intake Agent: No participant data, no customer financial data. First-name-last-initial rule enforced in identity prompt. Aug 17 2026 Atlassian training opt-out applies — confirm SyF tier setting before pilot rollout.
  • ⚠Synthesis Agent: Touches participant video + transcripts. HARD GATE — SyF legal must confirm platform consent covers AI synthesis. 30-day Atlassian retention on inputs/outputs — verify with bank data governance.
  • ✓No customer financial data: All research is prototype-based (Figma designs), not live customer PII. Clean from regulated-artifact perspective.
  • ✓Confluence write blocked: Already known from Rooster. All agent outputs exit as text; designer pastes into Figma/Google Doc.
🗂️

File Map

memory/drawers/day-job/jaime-rovo-agent-discovery.md — discovery prompt + project status

memory/drawers/day-job/jaime-rovo-discovery-output.md — Jaime's raw markdown export (verbatim)

memory/drawers/day-job/jaime-rovo-agents/README.md — portfolio build order + open-question triage

memory/drawers/day-job/jaime-rovo-agents/intake-agent.md — Phase 1 build spec

memory/drawers/day-job/jaime-rovo-agents/synthesis-agent.md — Phase 2 build spec (gated)

memory/drawers/day-job/jaime-rovo-agents/quarterly-digest-agent.md — Phase 3 deferred sketch

memory/tools/rovo-knowledge.md — ROVO platform reference

← Day Job drawer SyF Build Strategy Rooster Setup Rooster Mind Map