1 / 13
↑↓ Navigate F Fullscreen O Overview

AI for Project Managers

Code got cheap. Specs got expensive.
That's your career.

Product Owner Solution Builder

Most of you knew me when I was a Product Owner. A year ago I would have laughed at the idea of being where I am now. Then I started using AI seriously.

I know some of you have tried Copilot, gotten something garbage back, and concluded the hype is hype. I get it. I'm not going to tell you AI is magic — I'm going to tell you it's a coworker who needs onboarding.

Advance

Your BRD looks done. Marie has ten notes.

Two minutes. All ten obvious in hindsight.

  • 1Senior PM · 12 years tenure
  • 2Came up through stations operations
  • 3Plain English, no buzzwords
  • 4"Who's accountable?" before "what's the scope?"
  • 5Imagines specs read at 4am by a tired supervisor

Here's what I sent her.

A real-shaped BRD. Already through one round of internal review.

Cowork · BRD draft
workspace · driver-app
BRD · Driver app: smarter address-issue handling
v0.3 · self-reviewed · ready for stakeholder review
1. Background
Drivers report bad delivery addresses through manual escalation today. The flow is slow, inconsistent, and depends on the station supervisor relaying the issue to the address-correction team via email or radio.
2. Goals
Reduce address-related delivery failures by introducing an in-app reporting flow. Drivers should be able to flag a bad address in two taps and route it to the correction team automatically.
3. Scope
In-app reporting form, validation against the address-validation API, automatic routing to the address-correction team queue, status visibility for the driver.
> ask marie-review ⏎ SEND

Ten things I missed.

Cowork · BRD draft
workspace · driver-app
BRD · Driver app: smarter address-issue handling
v0.3 · sent for review · 2:14 ago
> ask marie-review SENT
marie-review · 10 catches
#1 · Hand-waving
"You wrote 'reduce address-related delivery failures.' By how much? In what window? Compared to what? Give me a number I can hold someone to."
#2 · Accountability
"You named the address-correction team. You didn't name a person. At 4am on a Tuesday when a driver is parked in a snowstorm waiting on a fix, who picks up the page?"
#3 · Black Friday
"This assumes normal volume. Drivers report 3-4x the address issues during peak. Has the correction team staffed for the surge? What's the queue depth at hour 6 on Black Friday?"
#4 · Feedback loop
"How does the driver know their report was received? You've designed the input. You haven't designed the response. A driver who reports a bad address and hears nothing will stop reporting."
#5 · Connectivity edge case
"Drivers in station dead zones — basements, loading docks, rural stops — have spotty signal. Does the report queue locally and sync later, or do we lose it? At 6am in the rain, that answer matters."
#6 · Sign-off gap
"At 4am, the night-shift supervisor is the one fielding driver complaints when the in-app flow breaks. Have you walked them through this? Have they signed off? They're not in your stakeholder list."
#7 · Implicit dependency
"This depends on the address-validation API. Has the platform team confirmed it handles the new request volume? Their last capacity review was October — what's changed since?"
#8 · Stakeholder gap
↳ Consulted: cx-review skill (Sarah Chen, Customer Experience)
"Customer Service isn't on your stakeholder list. They handle 18% of address-related calls — not drivers. They'll feel the rollout before driver ops will. Worth pulling Sarah in before scope #2 closes."
#9 · Compliance
↳ Consulted: legal-review skill (Mike Patel, Legal)
"Driver-reported addresses include PII — recipient names, building access codes, occasionally gate codes. Your data-retention policy on the report payload isn't specified. Mike will block this in review. Easier to head off now."
#10 · Cross-team upside
↳ Consulted: sustainability-impact skill (Lena Okafor, Sustainability)
"Reducing wrong-address incidents trims the wasted miles drivers run on bad routes. That's CO2 the sustainability team is currently measuring quarterly. 15-min sync with Lena — your project might land in her year-end report."
10 catches · 2:14 elapsed
I didn't program Marie. I described her.

She's a coworker, not a prompt. She remembers what she's looking for. Every time.

You've been writing prompts for years.
You called them specs.

BRDs, PRDs, user stories — they aren't going obsolete. They're the vocabulary AI was built to listen to.

What PMs already do BRDs · PRDs · user stories
acceptance criteria · specs
The overlap Same words.
New audience.
How AI works best Structured context · clear roles
explicit constraints · examples

Marie reviews specs.
The same pattern ships code.

Station One — live visibility for stations — has been in pilot for 2 months. The thing that closed the gap wasn't learning to code faster. It was learning to spec better.

BUILD ARC · STATION ONE JANUARY Prototype first working version TODAY In Pilot running in 6 stations MID-JUNE Prime Days v1.0 ready for peak

Every tool holds a piece.
None of them hold the picture.

VP · Outlook · 7:42am
Status on EPIC-247?
"I haven't seen a written update in two weeks — should I be concerned about the sortation rollout?"
Mon · unread
Jira · DPK-1142
Blocked · 6 days
Awaiting handheld vendor response. No escalation logged.
Last update Tue
Teams · #stations-platform
@you · 14 mentions
"Quick one — can you confirm the SLA cut-over is still Friday?"
Fri · 17:51
Confluence · sprint review
Scope change flagged
"Eng raised concern on delivery window — see notes section 3."
Fri · 16:08
Jira · 47 unread
Tickets need triage
5 high · 11 medium · 31 low — priority unclear from titles alone.
Backlog
Calendar · in 25 min
PMO sync
Weekly status round-table. Your draft email: still blank.
09:00 · Teams
Outlook · stakeholder thread
Re: re: re: re: timing
11 replies overnight from finance, ops, and legal. Subject changed twice.
Sat 22:14

The answer you need exists. It's just scattered across apps that don't talk.

Four tools. One answer.

Cowork · ad-hoc query
workspace · driver-app
> what's the real status of the driver address-issue feature? SENT
cross-tool synthesis · 4 sources
jira
↳ Consulted: 3 open tickets, project DPK
"DPK-1142 has been blocked 6 days waiting on the handheld vendor — no escalation logged. Last status update: Tuesday. Sprint velocity dropping."
outlook
↳ Consulted: stakeholder thread "Re: re: re: timing"
"VP-level thread Saturday: 11 overnight replies from finance, ops, and legal. Subject changed twice. Tone is sharpening — they're not waiting for Monday."
confluence
↳ Consulted: last sprint review · section 3
"Sprint review flagged a scope change on the delivery window. Eng raised concern. The Jira tickets haven't been updated to match — there's a quiet drift."
teams
↳ Consulted: #stations-platform · Friday 17:51
"Engineering lead flagged a dependency risk Friday afternoon. The thread didn't loop you in. Three of your stakeholders are silent watchers on it."
4 sources · synthesized in 1.2s

You don't build this.
You enable it.

Cowork ships with three drawers. The connectors I need are in one. The skills I'll write are in another. Someone already built the third.

This is the part Copilot can't do — Copilot reads one app at a time.

Cowork · Monday Morning Briefing
Last refresh · Mon Apr 6 · 7:58 am

Jon's brief — week of Apr 6

Next refresh · Mon Apr 13 · 8:00 am
Top 3 wanting your attention
Outlook VP · Mon 7:42 am
Asking about EPIC-247.
Last written update: 14 days ago. Slipping out of his mental model.
Jira DPK-1142 · blocked 6d
Handheld vendor unresponsive.
Vendor SLA breached. No escalation logged.
Teams #stations-platform · Fri 17:51
@you-mentioned. Thread moved on without you.
3 silent watchers — including Sarah Chen.
Initiative health
EPIC-204 · Driver app refreshOn track. Sprint 7 closes Fri.
EPIC-247 · Sortation rolloutVP asking — comms gap.
EPIC-251 · Address-issue handlingDPK-1142 blocked, at risk.
EPIC-218 · Returns flowQuiet — 0 risks flagged.
Quietly stuck
DPK-1098 No movement 9d · "Pending design review"
DPK-1087 No movement 12d · "Pending stakeholder feedback"
✎ Draft for your 9:00 am PMO sync
Driver app refresh on track for Sprint 7 close (Fri). Sortation rollout (EPIC-247) — comms gap noted, written update going out today. Address-issue handling at risk: DPK-1142 blocked 6d on handheld vendor, escalating to procurement this morning. Returns flow steady — no risks raised this week.
Cowork · Monday Morning Briefing
Last refresh · Mon Apr 13 · 7:58 am

Jon's brief — week of Apr 13

Next refresh · Mon Apr 20 · 8:00 am
Top 3 wanting your attention
Jira DPK-1142 · blocked 13d ↑
Procurement escalation made no movement.
Vendor SLA 2× breached. This is now a contractual issue.
Confluence Sprint review · Sat 16:08
Eng raised a possible 1-sprint slip.
Delivery window for EPIC-251 may need to slide. Decision needed by Wed.
Calendar Wed 14:00 · invite received
PMO sync moved to Wednesday.
Conflicts with your QBR prep block. Two of five attendees are tentative.
Initiative health
EPIC-204 · Driver app refreshOn track. Sprint 7 closes Fri.
EPIC-247 · Sortation rolloutVP signed off on last week's update.
EPIC-251 · Address-issue handlingDPK-1142 still blocked, slip risk.
EPIC-218 · Returns flowQuiet — 0 risks flagged.
Quietly stuck
DPK-1087 No movement 19d ↑ · Still pending stakeholder feedback
DPK-1156 No movement 8d · "Pending budget approval"
✎ Draft for your 9:00 am PMO sync
Driver app refresh tracking for Sprint 7 close (Fri). Sortation rollout (EPIC-247) — VP signed off; calm. Address-issue handling still at risk: DPK-1142 now 13d blocked, escalating to legal counsel today. Sprint review flagged a possible 1-sprint slip on EPIC-251 — meeting Eng lead Wed AM. Returns flow steady.
Cowork · Monday Morning Briefing
Last refresh · Mon Apr 20 · 7:58 am

Jon's brief — week of Apr 20

Next refresh · Mon Apr 27 · 8:00 am
Top 3 wanting your attention
Jira DPK-1142 · vendor responded ✓
Vendor unblocked. New dependency surfaced.
Driver app team needs an integration fix. ETA 5 days.
Outlook Customer · Sun 16:22
Strategic account flagged returns flow issue.
Tone is constructive but they cc'd their CFO. Reply expected by EOD.
Jira DPK-1201 · Sev-2
New defect on Sprint 7 release.
QA caught a regression in Driver app. Triage needed today.
Initiative health
EPIC-204 · Driver app refreshSprint 8 defect to triage.
EPIC-247 · Sortation rolloutCalm, no movement needed.
EPIC-251 · Address-issue handlingVendor unblocked, new dep being scoped.
EPIC-218 · Returns flowCustomer escalation; reply due.
Quietly stuck
DPK-1087 No movement 26d ↑ · Stakeholder feedback never came
DPK-1156 No movement 15d · Pending budget approval
✎ Draft for your 9:00 am PMO sync
Vendor unblocked DPK-1142 — new dep with Driver app team, 5d ETA. Sprint 7 closed; Sprint 8 has a Sev-2 defect to triage. Returns flow customer escalation needs an acknowledgment by EOD. Sortation rollout calm. EPIC-251 path to green is now visible.
Cowork · Monday Morning Briefing ● Now
Last refresh · Mon Apr 27 · 7:58 am

Jon's brief — week of Apr 27

Next refresh · Mon May 4 · 8:00 am
Top 3 wanting your attention
Jira DPK-1142 · ✓ closed · 21d e2e
Address-issue handling shipped.
Vendor relationship reset. Worth documenting the escalation path so it's faster next time.
Outlook Customer · Fri 11:04
Customer wants a joint root-cause meeting.
Constructive tone. They're proposing this week or next.
Calendar In 18 min · 09:00
PMO sync today — VP attending.
Briefing email below is ready. Skim it.
Initiative health
EPIC-204 · Driver app refreshDefect resolved. Sprint 8 underway.
EPIC-247 · Sortation rolloutSteady. No risks.
EPIC-251 · Address-issue handling✓ Shipped. Closing out.
EPIC-218 · Returns flowCustomer engaged constructively.
Quietly stuck
DPK-1087 No movement 33d ↑ · Recommend: kill or revive
✎ Draft for your 9:00 am PMO sync
Strong week. DPK-1142 closed (21 days end-to-end, retro on the calendar). Sprint 7 wrapped; Sprint 8 underway with the Driver app defect resolved. Customer returns escalation engaging constructively — joint root-cause meeting being scheduled. EPIC-247 sortation rollout green. No critical risks.
3 status emails auto-drafted this month · 100% sent rate
Cowork · Monthly Briefing ▦ Monthly
Last refresh · Thu Apr 30 · 8:12 am

Jon's monthly — April 2026

Next refresh · Sat May 31 · 8:00 am · 4 weekly snapshots synthesized
Initiative trajectory · 4-week arc
EPIC-204 · Driver app refresh Sprint 7 closed. Sprint 8 underway, 1 defect resolved.
EPIC-247 · Sortation rollout Comms rhythm restored. VP signed off Apr 13.
EPIC-251 · Address-issue handling Closed Apr 23 · 21 days end-to-end.
EPIC-218 · Returns flow Customer escalation engaged constructively.
Stories that resolved
DPK-1142 closed · 21d end-to-end
Sprint 7 wrapped · Apr 24
EPIC-247 unstuck · Apr 13
Customer returns escalation · Apr 27
Patterns this month
↳ vendor risk
73% of your top-3 attention items this month involved external vendor dependencies. Average response time: 8.2 days (industry benchmark: 3).
↳ silent stakeholder
Sarah Chen appeared as a silent watcher on 3 of 4 Fridays. Recommend looping her in actively before next month's review.
↳ stale ticket
DPK-1087 has been stuck for 33 days — same status, no movement. Recommend: kill or revive.
Heading into May
🎯
Peak season prep — Prime Days mid-June (6 weeks out) EPIC-204 polish needed. Driver app load test scheduled May 12.
SAP migration scoping starts May 5 Capacity model needs your input by Friday.
📊
Sortation rollout enters Phase 2 3 new stations onboarding · comms cadence already drafted.
April scoreboard
4/4
briefings refreshed
100%
status emails sent
11/12
attention items acted on
~5.5h
triage time saved

This is a Cowork artifact.
It runs on a schedule.
The data is always fresh.

Artifact
Live artifacts
+
Schedule
Every Monday 8am
+
Connectors
Jira · Confluence · Teams · ∞
=
Fresh brief
every week, automatically

Memory knows you.
Skills know how.

Memory is what Cowork remembers about you — your role, projects, how you work. Skills are what Claude knows how to do. And unlike memory, skills travel.

✦ Memory Cowork feature
What Claude knows about you
  • Your role + responsibilities
  • Projects, tools, teammates
  • How you like to work
Personal · automatic · stays with you
⚙ Skills
What Claude knows how to do
  • Marie's BRD review (you saw it)
  • Status report ghostwriter
  • Anything you'd hand a coworker
Shareable · repeatable · packageable

Marie wasn't sitting in my memory.
She was packaged.

A skill is a folder of instructions, templates, and rules. Claude only loads what's needed, when it's needed.

Progressive disclosure in skills Three stacked tiers showing how skills load progressively: name and description always loaded, the SKILL.md body loaded when relevant, and deep references and templates loaded only on demand. Progressive disclosure in skills A skill loads in layers — only what's needed, when it's needed TIER 1 · ALWAYS LOADED Name + one-line description "kickoff-package — generates a project launch package" TIER 2 · LOADED WHEN RELEVANT SKILL.md body — the recipe Workflow, instructions, examples Read when Claude detects a matching request TIER 3 · LOADED ON DEMAND References, templates, scripts BRD template, Marie's full persona, risk register format… Only opened when the recipe explicitly calls for them Can be huge — doesn't pollute context until needed

You don't need to build from scratch.

claudemarketplaces.com
Curated community · ~4,200 skills
claudeskills.info
~140 free skills · strong quality bar
skillhub.club
~7,000 community skills
deanpeters/PM-Skills
PRDs · problem framing · prioritization-advisor
phuryn/pm-skills
Chained workflows: /discover, /summarize-interview
alirezarezvani/claude-skills
Atlassian-native — Scrum Master, Jira Expert

Skills → what Claude knows. MCPs → what Claude can reach. Different things.

Explained the same thing
3 times this week?
That's a skill.

The Marie Critique
Adversarial spec review. Quotes hand-wavy language back at you. The antidote to sycophancy. Differentiator vs. Copilot.
🌩
Black Friday Stress Test
Walks any change proposal through peak conditions — 3x volume, short staffing, weather, handhelds dropping. Industry-native by design.

Marie was the warm-up.
Here's the bench.

Sometimes the missing input is a perspective, not a piece of data.

SE
Skeptical Engineer
"Show me the edge case you haven't thought of yet."
Spec audit
ST
Station Supervisor · 3am
"Looks good on paper. What breaks when the scanner drops signal?"
Ops reality
D47
Driver · Route 47
"I have 4 seconds per stop. Does this work with one hand?"
Last mile
DM
Dispatcher Mid-Storm
"Three routes just went down. What's your plan B?"
Crisis mode
CS
Customer Service Agent
"A customer is calling right now. What do I tell them?"
Frontline
SF
Shipper Finance Lead
"Where's the cost impact? I don't see it in the BRD."
P&L lens
CFO
CFO
"Give it to me in three lines. What's the payback period?"
Exec brief
M
Marie
"Who's accountable? I didn't see that in your spec."
PM critique

Save these as personas inside a skill. Now they're always available — without a meeting.

Tip 01 / 05
Watch your
context.

Even with memory, a single thread degrades. Long brainstorms drift. Start fresh threads more often than feels natural — and let memory carry what matters between them.

Response quality · across a long thread
Context start Context degraded ↘
Tip 02 / 05
Don't demand.
Let it interview you.

Details aren't the problem — demands are. Same context, different door: invite questions instead of asking for output.

❌ Output-only ask
Cowork · PRD draft

"Driver app v2 — March 15, team Marie/Sarah/Mike, address-issue handling, risks A through G, deps with sortation + dispatch, success = adoption + CSAT, plus training, rollback, peak prep — please write the PRD."

↳ Door closed. PRD comes back generic. You rewrite for an hour.

✓ Context + interview
Cowork · PRD draft

"Driver app v2 — March 15, team Marie/Sarah/Mike, address-issue handling, risks A through G, deps with sortation + dispatch, success = adoption + CSAT, plus training, rollback, peak prep. Before we draft — what's missing?"

Claude → Two things: rollback per-feature or all-or-nothing? And — has Marie weighed in on peak-day exception flow yet?

↳ Same context. Door open. The gaps surface before the draft.

Tip 03 / 05
Set Sonnet as default.
Opus for the hard stuff.

Sonnet handles 95% of PM work. Faster iteration = faster learning. Opus is the senior partner you call in for deep reasoning — not your daily driver.

Model
Claude Sonnet 4.6
Fast · capable · the workhorse
Recommended
Claude Opus 4.6
Deep reasoning · slower · save for the hard ones
Power tool
Claude Haiku 4.5
Cheapest · quick lookups · fine for tiny tasks
When to call Opus in:
  • Multi-stakeholder synthesis with conflicting inputs
  • Long architecture or strategy reasoning
  • When Sonnet keeps missing the same nuance
Otherwise: Sonnet. Always.
Tip 04 / 05
Clear and polite.

Vague-but-polite asks get vague-but-polite work back. Specificity is respect — for Claude, and for the coworker you'd give the same task to.

❌ Polite, not clear

"hey claude, could you maybe help me write something about the driver app rollout? thanks!"

↳ You get something. You rewrite it. You wonder if AI is overrated.

✓ Clear and polite

"Write a 1-pager for ops leadership on the driver app v2 rollout — what changes Friday, what doesn't, what to tell drivers. Plain language. Keep it warm."

↳ Ships in one round. Politeness costs nothing. Specificity buys everything.

Tip 05 / 05
Project instructions.
Healthy AND lean.

Every project has rules you don't want to re-explain. Put them in the project instructions — Cowork calls it Folder instructions, Claude Code calls it CLAUDE.md, same thing. Then keep it ruthlessly short.

Folder instructions

Use this to give Claude instructions for working in this folder.

# Driver app v2 - Push back when something doesn't add up. - We say "station," not "depot." - Marie tone for spec review; CFO tone for exec briefs. - Don't paste long specs into chat — link to /specs. - Ask who the audience is before drafting.

Belongs

  • Standing rules
  • Terminology you correct twice
  • Tone defaults by audience
  • Where things live

Doesn't belong

  • This week's tasks
  • Today's facts
  • Anything that changes weekly

Rule of thumb: under 20 lines. Trim quarterly. Long is the same as no rules.

Four things that bite.
Four things you can plan for.

🎲
Hallucinations
It will sometimes make things up — confidently. Numbers, citations, specifics. It doesn't know what it doesn't know.
→ Verify outputs. Trust but check.
🪞
Sycophancy
It will agree with you when it shouldn't. Validate your bad ideas. Tell you what you want to hear. This is real and well-documented.
→ Disagree with it. Ask: "what's wrong with my plan?"
🔒
Privacy
Customer data, PII, operational details — don't paste them into public models. Company-approved tools don't train on your data, safe for confidential info.
→ Unsure? Ask cybersecurity.
✍️
You're still accountable
It surfaces options, critiques, drafts. The decision stays with you. Marie can flag ten catches — your name's still on the BRD.
→ Read it like you'll have to defend it.
Don't try everything.
Build a Marie this week.

Pick one. Do it once. That's the whole assignment.

Default · easiest start
Build a Marie.
Pick a senior reviewer on your team — real or composite. Describe how she reads a draft, what she insists on, what makes her sigh. Then send your next BRD past her.
No connectors. No code. Just a description. You already know how she thinks.
Or — go cross-tool
Ask the real status.
Pick your top initiative. Ask Cowork what's actually happening with it — let it pull from your tools instead of you stitching it together.
Or — go scheduled
Schedule a Monday brief.
One artifact. One schedule. Your inbox a little lighter every Monday morning.
You won't figure this out alone.
Neither did I.

Two surfaces for the part that comes after the talk.

The community
Engage.
Our internal Anthropic community. Where we trade what worked, what bit us, and what we built. Not a help desk — a workshop. I'm in there too, as a participant.
Built a Marie that caught something good? Brag. Borrow. Steal.
The packet
Coming this week.
My skills doc · the context-bloat doc · the PM marketplaces and repos worth borrowing from. Shared with the meeting files.
Code got cheap.
Imagination got expensive.
Specs got you here.
Questions get you further.