Agentic Team Flow

Hybrid workflow combining human decision-making with AI agent execution. Humans control strategy; agents handle operations.

Overview

Pattern Description

This pattern divides responsibilities:

• Humans: Architecture decisions, code review, release approval
• Agents: CI/CD, automated checks, notifications, merges, documentation

Automation Level: ~80% Human Intervention: Strategic decisions only

Participants

Branch Structure

main (production)
 └── develop (integration)
      └── feature/* (human-created)

Workflow Phases

1. Feature Development (Human)

# Developer creates feature branch
git checkout -b feature/add-auth develop

# Developer implements changes
git add .
git commit -m "feat: add authentication"

# Developer pushes and creates PR
git push origin feature/add-auth
# → Create PR via GitHub/GitLab UI

2. Automated Review (Agent)

When PR is created, agent automatically:

┌──────────────────────────────────────────────────┐
│              Agent Review Pipeline               │
├──────────────────────────────────────────────────┤
│  1. Run test suite                               │
│  2. Check code style (lint, format)              │
│  3. Build project                                │
│  4. Security vulnerability scan                  │
│  5. Generate review summary                      │
│  6. Identify auto-fixable issues                 │
│  7. Apply auto-fixes (formatting, imports)       │
│  8. Notify reviewers                             │
└──────────────────────────────────────────────────┘

Agent Comments on PR:

## 🤖 Automated Review Summary

### ✅ Checks Passed
- Unit tests: 142 passed
- Coverage: 84% (+2%)
- Build: Success (32s)

### 🔧 Auto-Fixed
- Formatted 3 files
- Sorted imports in auth.py

### ⚠️ Suggestions
- Consider adding error handling in `login()` (line 42)
- Test coverage for `logout()` is below threshold

### 👥 Ready for human review
@team-lead @senior-dev

3. Human Review (Reviewer/Lead)

Reviewer focuses on:

• Architecture alignment
• Business logic correctness
• Edge cases and error handling
• Security considerations
• Performance implications

Review outcomes:

• ✅ Approve → Agent proceeds to merge
• 📝 Request changes → Developer addresses feedback
• 💬 Comment → Discussion without blocking

4. Agent Merge

After human approval:

1. Agent merges PR (squash or rebase per config)
2. Deletes feature branch
3. Updates release notes
4. Notifies team via Slack/Discord

5. Release Cycle (Weekly)

Team Lead                    Agent
    │                          │
    │──── Trigger release ────>│
    │                          │──── Create release branch
    │                          │──── Compile changelog
    │                          │──── Bump version
    │<── Release ready ────────│
    │                          │
    │──── Review & approve ───>│
    │                          │──── Merge to main
    │                          │──── Create tag
    │                          │──── Deploy
    │<── 🚀 Deployed ──────────│

Communication

Slack/Discord Notifications

GitHub/GitLab Comments

Agent posts structured comments with:

• Test results summary
• Coverage diff
• Build logs (on failure)
• Auto-fix details
• Suggested improvements

Weekly Summary Email

📊 Weekly Development Summary

PRs Merged: 12
Tests Run: 1,847
Coverage: 84% (+3%)
Releases: 1 (v1.2.0)
Issues Resolved: 8

Top Contributors:
1. @developer-a (5 PRs)
2. @developer-b (4 PRs)
3. @developer-c (3 PRs)

Configuration

Agent Behavior

agent:
  auto_fix:
    formatting: true
    imports: true
    trailing_whitespace: true
  reviews:
    generate_summary: true
    suggest_improvements: true
    max_suggestions: 5
  merge:
    strategy: squash  # or rebase, merge
    delete_branch: true
    update_changelog: true

Approval Requirements

approvals:
  required: 1
  from_code_owners: true
  dismiss_stale: true
  require_review_from:
    - team-lead
    - senior-dev

Release Schedule

release:
  schedule: "0 10 * * 1"  # Monday 10:00
  require_approval: true
  approvers:
    - team-lead
    - product-owner

When to Use

✅ Good Fit:

• Teams of 3-10 developers
• Weekly release cycles
• Need audit trail for compliance
• Want AI assistance without full automation
• Balancing speed with oversight

❌ Poor Fit:

• Solo developers (overkill)
• Very large teams (need more structure)
• Continuous deployment requirements
• Teams resistant to automation

Metrics

Related Patterns

• Agentic Git Flows — Full automation
• Simplified GitFlow — Manual process

← Back to Git Flows Overview