How I'd build the AI operating system every product manager actually needs — not another integration, but the intelligence layer that makes every PM workflow faster, more defensible, and less exhausting.
A note on this case study
This is a speculative product case study — not a shipped product, but a fully worked-through design proposal. I wrote it because I kept running into the same friction at work: my tools don't talk to each other, and intelligence lives in my head instead of my system. This is how I'd solve it.
01 — Problem Framing
A product manager's job is to make good decisions faster than anyone else on the team. But the reality of most PM workflows is exactly the opposite: we spend an enormous portion of every week synthesizing information — reading Slack threads, pulling Jira status, skimming Confluence docs, reviewing customer feedback — before we can even begin to make a call.
The tools exist. The data exists. But the intelligence layer that connects them doesn't. Every PM is manually operating as the integration point between systems that were never designed to talk to each other.
That's the problem. Not a feature gap — a structural gap. The question is what to build to fix it.
How PMs spend their time (the honest version)
02 — Market Context
68%
of PMs report spending more than 20 hours per week on tasks they describe as "administrative"
Product Collective Survey, 2024
7–10
avg. number of tools in a senior PM's daily stack — none of which share context with each other
Industry average
$47B
projected AI productivity software market by 2028 — most of which is targeting developers, not PMs
Gartner, 2025
Why existing solutions fall short
Notion AI / Confluence AI
Helpful for document drafting, but siloed within a single tool. Doesn't know about your Jira board or your Slack threads.
"AI in Jira" features
Summarizes tickets. Doesn't understand roadmap context, team dynamics, or strategic tradeoffs.
ChatGPT / Claude (generic)
Powerful but stateless. You paste context in, it responds, and tomorrow it remembers nothing about your product.
PM-specific AI tools (ProductBoard AI, etc.)
Feature-level AI on top of a single tool. Doesn't address the cross-tool intelligence gap.
03 — User Personas
Sarah
Junior PM · Series B Startup
Writes PRDs from scratch and receives feedback only after she's gone deep. Unsure of her instincts. Spends hours formatting documents that should take minutes. Doesn't have a senior PM to shadow.
Core need
A thinking partner that catches blind spots before the PRD review
Marcus
Senior PM · Enterprise SaaS
Manages three workstreams and eight stakeholders. Loses an hour each morning just re-orienting. Context-switches 15+ times per day. His biggest productivity leak is async thread management.
Core need
Fast context recovery and auto-generated stakeholder briefs
Linda
Head of Product · 200-person company
Manages a team of six PMs across two products. Can't see what's happening without weekly status meetings that take half a day. Strategic decisions are bottlenecked on information she never has in the right format.
Core need
Portfolio-level visibility without the meeting overhead
Common thread: All three personas are drowning in information that's technically available but practically inaccessible. The solution isn't more information — it's intelligence that synthesizes and surfaces what matters, in the format they need, when they need it.
04 — Options Evaluation
AI copilot inside existing tools
Notion AI, Confluence AI, Jira AI
Standalone PM AI assistant
Chat interface for PM questions, no integrations
Connected intelligence layer
SelectedAI layer that reads across Jira, Notion, Slack, and customer feedback to synthesize cross-tool context
Opinionated PM app with AI-native design
Replace the stack with a new tool built for AI-era PM workflows
Option 3 (connected intelligence layer) is right but hard to build. Option 4 (replace the stack) is the dream but requires behavior change that rarely happens at scale. The MVP strategy is to win on a single workflow first — something that replaces one hour of a PM's week so convincingly they can't imagine going back. Then expand the connection surface from there.
05 — The Product
Four core modules. Each one replaces a specific type of PM friction.
Document Intelligence
AI reads PRDs, meeting notes, and customer feedback and surfaces patterns. "Three customers this week mentioned they can't find the export button" — without you having to read every ticket.
Key behavior: Proactively flags tensions between your PRD and recent customer feedback — before sprint planning, not after.
Decision Support
Before you ship a feature, the system runs an automated check: does this conflict with a prior product decision? Does the data support the hypothesis? Have similar bets failed elsewhere?
Key behavior: Surfaces your own historical decisions so you're not repeating debates you already had six months ago.
Stakeholder Briefings
Auto-generates weekly status updates from Jira and Linear state, formatted differently for engineering (technical detail), leadership (business impact), and cross-functional partners (dependencies only).
Key behavior: You review and approve — you don't write from scratch. Every stakeholder gets the format they actually process.
North Star Tracker
Live dashboard mapping your active roadmap initiatives to their declared metrics. Flags when a shipped feature hasn't moved its stated metric in 30 days — so nothing ships and disappears into the void.
Key behavior: Creates accountability between the story you told in the PRD and the results you're seeing in production.
06 — North Star & Metrics
North Star Metric
% of PM work output with an AI-assist component
Target: 60% within 6 months of launch. This measures product depth — not just activation, but whether PMs are relying on the system for their actual work, not just trying it once.
Leading Indicators
Lagging Indicators
Counter-Metrics
07 — Wireframes
The guiding principle: every screen should feel like talking to someone who already read your Jira board.
View 1 — Morning Briefing (Daily digest)
What needs your attention today
Sprint conflict detected
"Data Export v2" scoped for this sprint — backend estimate changed from 3 pts → 8 pts.
Matched to: Q2 Roadmap · Data Platform initiative
Customer signal — 3 new tickets
Theme: Users can't find bulk edit on mobile. Matches open initiative: Mobile UX Refresh (Q2)
3 tickets · 2 severity-3 · 1 severity-2
Stakeholder briefing ready to review
Auto-generated from 7 days of sprint activity · Recipients: Eng Lead, VP Product, CS
Today's focus
Review sprint re-scope options for Data Export v2
Approve stakeholder briefing before 11am standup
PRD review: Notification Preferences (3 comments)
View 2 — PRD Draft Mode with Decision Support
Problem Statement
AI Assist
Prior Decision
Similar scope explored Q3 2024 (Notification Audit PRD). Delayed due to event service dependency.
That dependency shipped Dec 2024
Risk Flag
Mobile notification UX flagged by 3 users this week. Consider mobile-first variant.
08 — Phased Rollout
Phase 1 — Stakeholder Briefing Generator
MVPConnect to Jira/Linear. Auto-generate a weekly status brief from sprint activity. PM reviews, edits, and sends. One integration. One workflow. Saves 45–90 minutes per week immediately.
Why start here: It's the easiest PM behavior to observe (everyone writes status updates), the value is immediately legible, and the integration surface is narrow. If we can't win here, we can't win anywhere.
Phase 2 — Document Intelligence + PRD Assist
ExpansionConnect to Notion/Confluence. Read existing PRDs, past decisions, customer feedback. Surface relevant context inline while PMs write. The system learns your product and stops letting institutional knowledge disappear when people leave.
Why second: Requires more trust. PMs need to have seen Phase 1 work reliably before they'll let AI near their PRDs.
Phase 3 — Full Intelligence Layer
VisionCross-tool intelligence across Jira, Notion, Slack, customer feedback, and your metrics stack. Morning briefing surfaces the three things that matter today. North Star Tracker flags when shipped features aren't moving their stated metrics. The PM's full working memory — externalised and automated.
The goal: Every PM on the platform operates with the institutional knowledge of someone who's been in the role for 5 years — regardless of tenure.
09 — Reflection
The real risks
The hard product question
Why I wrote this: The PM productivity gap is real and personal. I built parts of this system for my own workflows — internal agents that read customer tickets, draft summaries, and flag anomalies in product data. The technology works. What's missing is the opinionated product layer that makes it usable by PMs who aren't also AI builders. That's the gap. That's the product.