Practical AI workflows for Delivery Program Managers — built by someone who has run real standups, written real escalations, and sat through more sprint retros than they'd like to admit.
🌐 Live site: nikhil-thomas-a.github.io/pm-ai-hub
Most AI tools are built for developers, by developers. The PM experience is an afterthought — generic prompts, abstract use cases, no grounding in delivery reality.
This is different. Every prompt, tool, and case study here was designed for the actual problems a Delivery PM faces: mid-sprint escalations, stakeholder reporting under pressure, delivery risk conversations with leadership, and retros that need to uncover root causes without turning into blame sessions.
Copy-paste prompts for the three highest-leverage PM pain points:
| Category | What It Solves | Time Saved |
|---|---|---|
| Stakeholder Reporting | Turns raw sprint data into exec-ready updates | ~45 min/week |
| Escalation & Incidents | Drafts escalations and post-mortems under pressure | ~30 min/incident |
| Sprint & Delivery Tracking | Spots risks early, surfaces patterns across retros | ~1 hr/sprint |
Each prompt includes the exact text, difficulty level, time estimate, pro tips, and a step-by-step guide.
Live AI-powered tools PMs can use right now:
Slack Summariser Reads a Slack channel dump and extracts:
- Active blockers (with owner if mentioned)
- Decisions made
- Action items with due dates
- Escalation candidates
Powered by Claude API + Slack MCP integration.
Real delivery problems, exact prompts used, measurable outcomes.
Situation: Sprint day 6. Three stories blocked on a third-party API. Stakeholder expecting demo in 4 days.
Prompt used: PM AI Hub escalation template (30-second draft)
Output: Structured escalation email drafted in under 2 minutes, covering: impact summary, root cause, options considered, recommended path, and ask.
Time saved vs. manual drafting: ~25 minutes. More importantly — sent within 10 minutes of identifying the issue, not 2 hours later after agonising over wording.
Situation: 6-sprint retrospective data sitting in Miro. Need to find systemic patterns before quarterly planning.
Prompt used: Pattern extraction prompt from Playbook
Output: Claude identified 3 recurring themes across 6 retros (dependency delays, unclear acceptance criteria, insufficient testing time), ranked by frequency and severity.
Time saved: ~3 hours of manual synthesis → 8-minute prompt session.
| Layer | Technology |
|---|---|
| Frontend | React 18 + Vite |
| AI | Claude API (Anthropic — claude-sonnet) |
| Integration | Slack MCP |
| CI/CD | GitHub Actions → GitHub Pages |
| Hosting | GitHub Pages |
git clone https://github.com/nikhil-thomas-a/pm-ai-hub.git
cd pm-ai-hub
npm install
npm run devOpen http://localhost:5173
No API key needed for browsing — AI features require a Claude API key set as VITE_CLAUDE_API_KEY in .env.
- Risk radar — paste sprint data, get a structured risk assessment
- Retro analyser — upload retro notes, get recurring themes + action items
- AI training explainer — plain-English guide to SFT, RLHF, RLEF, and eval metrics (Pass@K, HumanEval) for PMs working with AI teams
- Delivery health score — weekly programme health score from Jira data
Nikhil Thomas A — Delivery Program Manager exploring the intersection of AI and programme delivery.