How to ship from issue to merge with the GitHub Copilot app | demo

Key Concepts

• GitHub Copilot App: A centralized workspace for managing the entire Software Development Life Cycle (SDLC).
• Work Trees: A Git feature that allows multiple branches to be checked out into separate folders simultaneously, preventing context-switching overhead.
• Agentic Workflow: An approach where the AI agent drafts plans, implements code, and iterates based on human feedback.
• Agent Merge: An automated feature that monitors pull requests (PRs), handles review feedback, ensures CI passes, and merges code.
• Pick and Polish: A visual debugging/editing tool that uses screenshots to provide context to the AI, eliminating the need for manual selector descriptions.

---

1. The Problem: Context Switching

The video highlights the "cost" of multitasking in software development: the mental and technical effort of switching between branches, stashing changes, and remembering the state of multiple tasks. The GitHub Copilot app aims to eliminate this by centralizing all issues and PRs into a single interface.

2. Workflow Methodology: The "Session" Framework

The app organizes work into "Sessions," which provide isolation for different tasks:

• Centralized View: Issues and PRs are categorized into Active, Review Requested, and Done. Users can create custom sections using GitHub search queries.
• Isolation via Work Trees: When a session starts, the app checks the branch into a dedicated folder on the disk. This ensures that editing a file in one session does not affect others, removing the need for git stash or git checkout shuffles.
• Auto-Start Sessions: If enabled, the agent automatically drafts a plan and commits it to the work tree as soon as a session begins.

3. Human-AI Collaboration

The speaker emphasizes an outcome-focused approach rather than passive automation:

• Planning: The agent proposes a technical plan. The developer reviews and provides feedback in plain English.
• Implementation: The agent executes the code changes.
• Verification: Developers can run the application locally directly from the work tree to verify the agent's output.
• Pick and Polish: A specialized tool where the user clicks on a UI element; the app captures a screenshot and attaches it to the composer, allowing the AI to understand the visual context without manual input.

4. Agent Merge: Automating the PR Lifecycle

"Agent Merge" is presented as the solution to the bottleneck of "baby-sitting" PRs. It operates by continuously checking three criteria:

1. Review Threads: Have all comments been addressed?
2. CI Status: Is the build/test suite green?
3. Branch Integrity: Is the branch clean?

If the agent encounters a blocker, it notifies the developer; otherwise, it autonomously merges the PR. This allows developers to manage multiple PRs simultaneously without manual intervention.

5. Automations

The app includes built-in automations for repetitive tasks, such as:

• Triage: Automatically sorting and labeling new issues.
• Maintenance: Sweeping stale branches to keep the repository clean.

---

Synthesis and Conclusion

The GitHub Copilot app transforms the developer experience from a manual, branch-heavy workflow to an agent-orchestrated SDLC. By leveraging Work Trees for technical isolation and Agent Merge for lifecycle management, the developer shifts from being a "task switcher" to an "outcome manager."

Main Takeaway: The platform enables a significant increase in throughput—moving from shipping one PR per day to managing multiple PRs and issues concurrently—by allowing the AI to handle the implementation and maintenance details while the developer focuses on high-level intent and feedback.