Built by Monil Raval — Certified SAFe 6 POPM | Product Owner | Ex-AGCO/Fendt, Bosch | MBA Germany
AgentSprint is one of the first multi-agent AI systems purpose-built for Agile Sprint Planning. Describe a feature or product challenge, and 6 specialised AI agents — each with their own persona, expertise, and system prompt — collaborate in real time to produce a complete, sprint-ready backlog.
No other tool does this. Existing AI tools for product management generate user stories in isolation. AgentSprint simulates an entire agile team debating, estimating, risk-scoring, and reviewing — just like a real PI Planning session.
User Input
│
▼
┌─────────────────────────────────────────────────┐
│ ORCHESTRATOR AGENT 🎯 │
│ Receives input · Decomposes · Coordinates │
└─────────────────┬───────────────────────────────┘
│ delegates in parallel
┌─────────┼─────────┐
▼ ▼ ▼
┌────────┐ ┌────────┐ ┌────────┐
│PO Agent│ │Dev Agent│ │Risk │
│👤 │ │💻 │ │Agent ⚠️│
│Stories │ │Estimates│ │Risks │
└────────┘ └────────┘ └────────┘
│ sequential review
▼
┌────────────┐ ┌────────────┐
│Scrum Master│ │ QA Agent │
│🔄 DoR/DoD │ │🧪 Tests │
└────────────┘ └────────────┘
│
▼
┌─────────────────────────────────┐
│ SYNTHESISED SPRINT BACKLOG │
│ User Stories · Estimates · │
│ Risks · Tests · DoD │
└─────────────────────────────────┘
| Agent | Role | Output |
|---|---|---|
| 🎯 Orchestrator | Coordinates all agents, synthesises final plan | Sprint kick-off · Final summary |
| 👤 PO Agent | Product Owner perspective | 4 User Stories · Acceptance Criteria · Priorities |
| 💻 Dev Agent | Engineering perspective | Effort estimate · Tech stack · Dependencies |
| Risk management perspective | 3 Risks · Severity · Mitigations · Risk Score | |
| 🔄 Scrum Master | Process compliance | DoR check · DoD criteria · Blockers · Readiness score |
| 🧪 QA Agent | Quality assurance perspective | Test scenarios · Edge cases · Exit criteria |
Try it with one of these prompts:
- "Build a real-time PIM data quality dashboard for product managers to monitor and fix data errors across 4 global brands"
- "Add live charger availability and session pre-booking to a mobile app for EV drivers"
- "Create a fleet management portal with consolidated billing and cost centre allocation"
I spent 10 months as a Product Owner at AGCO/Fendt running SAFe PI Planning for 4 teams across global brands. Sprint planning consumed 2–3 days per sprint. The process was:
- Write user stories (PO)
- Get dev estimates (Dev Team)
- Identify risks (Risk Register)
- Check DoR compliance (Scrum Master)
- Write test scenarios (QA)
- Synthesise into a sprint plan (Everyone)
AgentSprint simulates steps 1–6 in under 60 seconds.
AgentSprint uses a sequential orchestration pattern with parallel delegation:
// Each agent has a unique system prompt (persona)
const AGENTS = {
orchestrator: { persona: "You are the Orchestrator Agent..." },
po: { persona: "You are the Product Owner Agent..." },
dev: { persona: "You are the Developer Agent..." },
risk: { persona: "You are the Risk Agent..." },
sm: { persona: "You are the Scrum Master Agent..." },
qa: { persona: "You are the QA Agent..." }
};
// Orchestration flow
async function runAgents(feature) {
await callClaude(AGENTS.orchestrator.persona, feature); // Kick-off
await callClaude(AGENTS.po.persona, feature); // User stories
await callClaude(AGENTS.dev.persona, feature); // Estimates
await callClaude(AGENTS.risk.persona, feature); // Risks
await callClaude(AGENTS.sm.persona, feature); // DoR/DoD
await callClaude(AGENTS.qa.persona, feature); // Tests
// Synthesise → Sprint Plan
}| Decision | Choice | Reason |
|---|---|---|
| Framework | Vanilla JS | Zero dependencies, runs anywhere, GitHub Pages compatible |
| Agent Communication | Sequential API calls | Simpler to debug, easier to follow for demo purposes |
| Model | Claude Sonnet claude-sonnet-4-20250514 | Best balance of speed and quality for agentic tasks |
| Persistence | In-memory | No backend needed, fully client-side |
agentsprint/
├── index.html # Complete app — single file, zero dependencies
├── README.md # This file
├── architecture.md # Deep dive into multi-agent design
└── examples/
└── sample-output.md # Example sprint plan outputs
git clone https://github.com/monilraval/agentsprint.git
cd agentsprint
open index.html # That's it. No npm. No install. No config.To enable AI generation, the app calls the Anthropic API directly from the browser. Add your API key in the fetch headers (for local testing only — never commit keys).
| Feature | AgentSprint | ChatGPT prompt | Jira AI | GitHub Copilot |
|---|---|---|---|---|
| Multiple agent perspectives | ✅ 6 agents | ❌ Single response | ❌ | ❌ |
| Dedicated Risk Agent | ✅ | ❌ | ❌ | ❌ |
| SAFe methodology built-in | ✅ | ❌ | ❌ | ❌ |
| DoR/DoD compliance check | ✅ | ❌ | ❌ | ❌ |
| Zero dependencies | ✅ | N/A | ❌ | ❌ |
| Open source | ✅ | ❌ | ❌ | ❌ |
| Built by a real PO | ✅ | ❌ | ❌ | ❌ |
- v1.1 — Export sprint plan as JIRA-compatible CSV
- v1.2 — Persistent sprint history across sessions
- v1.3 — Agent memory (agents reference previous sprint decisions)
- v2.0 — Real parallel agent execution using Web Workers
- v2.1 — SAFe PI Planning mode (multiple teams, Program Board)
PRs welcome. If you're a Product Owner, Scrum Master, or agile practitioner and want to improve the agent personas — open an issue or PR.
MIT — free to use, fork, and build on.
| linkedin.com/in/monil-raval | |
| Website | clarushorizon.com |
| monilraval@gmail.com | |
| SAFe Cert | ID: 76253775-6778 |
AgentSprint was built because the best way to demonstrate product thinking is to build the product. This is how I work: I don't just talk about agile — I ship it.
⭐ If this helped you, please star the repo — it helps other PMs find it.