Ship Code. Without Writing It.
summary
Demoed system takes a Jira ticket, reads project code with client, consults custom MCP protocol servers for Jira/Git/Confluence context, and raises a branch + PR with commit/branch naming and optional tests; NA10/NAT orchestrates agents for requirement parsing, design, coding and an RCA agent scans logs. Reports show end-to-end runs on production apps with roughly 20–30% failure rate for PRs today while Jira to PR automation handles mundane chores and leaves humans for review and architecture.
AI-written summary · about
chapters (22)
- 0:00Welcome
- 1:08Speakers set context
- 2:11LLMs to AI agents
- 5:27Multi-agent systems
- 10:22Introducing client
- 12:27Demo: client builds snake game
- 20:31How we use client in day-to-day development
- 21:59Solving LLM integration with MCP
- 25:32Demo: Jira to PR with MCP
- 28:32Tickets to autonomous PRs
- 29:56Demo: NAN automates Jira ticket
- 35:41Build agent in NAN
- 39:55Handling complex projects
- 41:18Humans in AI dev
- 42:23Key takeaways
- 43:25Tool choice mindset
- 44:08Q&A
- 44:52No-code for complex projects?
- 46:19AI test cases
- 47:42Client vs Cursor vs Copilot
- 50:57Fintech AI use cases
- 53:26Closing
related talks
16mShip Production Software in Minutes, Not Months
47mTales from the Dev Side: What Went Wrong, What We Fixe & What You Can Steal
30mHow do we code review at Slack
38mHow to Optimize Non-Coding Time
43mCode Review Tools
35mDevToDeploy: Integrate Your PM & CICD Tool For Systematic & Faster SW Delivery
36mAgentic CI/CD: from Pull Request to Production without the Paper Cuts
38mScaling Code Quality at Cisco | The Evolved AI-Driven SDLC