Workshops: Agentic Engineering Practices
Git Workflows for AI Coding Agents: Half-Day Workshop
Coding agents change many files quickly, and without a disciplined Git workflow that speed turns into tangled diffs, unreviewable pull requests and changes nobody can cleanly undo. In this half-day workshop your team sets up a Git workflow built for agents: isolated branches and worktrees, small commits, frequent save points and a reliable way back.
Who is this workshop for?
- Developers who use coding agents daily and want more control over their changes
- Tech leads who review agent-generated pull requests
- Teams running several agents or sessions in parallel on one repository
- Maintainers of repositories with strict history and review rules
- Release managers and DevOps engineers responsible for branch protection
What you will take away
- Give every agent task its own branch or worktree so parallel work does not collide.
- Have agents commit in small, meaningful steps with messages that explain the why.
- Use checkpoints, stashes and commits as save points and discard failed attempts in seconds.
- Keep agent pull requests small enough for a real review and make AI involvement transparent.
- Protect main branches with branch protection, required checks and limited agent permissions.
Agenda
Module 1
Git as the agent's safety net
- Why every agent session should start from a clean working tree
- Built-in checkpoints in Claude Code and Cursor compared with real Git commits
- What agents may do in Git and what stays with people, such as force pushes or rebasing shared branches
Module 2
Branches and worktrees
- One task, one branch: naming conventions for agent work
- Git worktrees for parallel agent sessions without extra clones
- Exercise: two agents, two worktrees, one merge
Module 3
Commits and pull requests
- Commit hygiene: small steps, conventional commits, co-author trailers
- Splitting a large agent diff into a reviewable stack of pull requests
- PR descriptions the agent drafts and a human verifies
Module 4
Undoing and recovering
- Revert, reset and the reflog as the last resort
- When the agent changed far too much
- Exercise: rescue a broken agent session without losing the good changes
Why Git discipline matters more with agents
A developer typically makes a handful of related changes and commits them. An agent may touch thirty files in a few minutes, mix a bug fix with an unrequested refactoring and keep going after its first mistake. Without a clear workflow, the useful part and the damage end up in the same messy working tree.
Git already has everything needed to control this: branches for isolation, commits as save points, revert for safe undo. The workshop turns these into team rules and agent instructions, so every agent works the same way, whether Claude Code, OpenAI Codex, GitHub Copilot or Cursor.
Rules you take home
Participants leave with a short Git policy for agent work that can go straight into your instruction files:
- Agents never commit directly to main or release branches
- One logical change per commit, tests passing before each commit
- Pull requests above an agreed size are split before review
- Agent involvement is visible in commits or PR labels
- Destructive Git commands require explicit human approval
Short, practical, in your repository
The workshop runs in your own repository or a sandbox copy, remotely or on-site, in English or German, and works with GitHub, GitLab, Bitbucket and Azure DevOps. We assume everyday Git skills and focus on the habits and recovery techniques that agents make necessary.
Frequently asked questions
Should agents be allowed to commit at all?
In their own branches, yes: frequent agent commits make rollback easy. Merging into protected branches stays a human decision through pull requests and required checks.
Are agent checkpoints enough as a safety net?
They are handy for quick undo within a session, but they do not replace Git. Checkpoints typically track only the agent's own file edits, not every shell command, and live locally in the tool; commits are shared, durable and reviewable.
How do we mark AI-generated changes?
Common options are co-author trailers in commit messages, PR labels or a section in the PR template. We help you choose a convention that fits your review and audit requirements.
Workshop
Get in touch
Want agent changes in your repository to be small, traceable and easy to undo?
Tell us roughly how many people will join and pick dates that work – we will get back to you within 24 hours with a proposal for content, format and schedule.
Prefer to talk first?
15 minutes, free and without obligation – with our AI Trainers and AI Code Engineers.
Book a call for free (opens in new tab) +49 761 519 751 50 (24/7)Request a training / workshop
Workshop: Git Workflows for AI Coding Agents: Half-Day Workshop · Half day (approx. 4 hours)
More workshops
- Workshops: Agentic Engineering PracticesSelf-Healing CI with AI Agents: Builds, Flaky Tests, Updates
- Workshops: Agentic Engineering PracticesTaking Over Vibe-Coded Prototypes: Workshop for Dev Teams
- Workshops: Agentic Engineering PracticesDesign Systems & Storybook with AI Coding Agents: Workshop
- Workshops: Agentic Engineering PracticesPair & Mob Programming with an AI Agent: Team Workshop
- Workshops: Agentic Engineering PracticesAgent-Ready User Stories: Scrum & Kanban with Coding Agents
- Agentic Coding WorkshopsAgentic Coding Training Curriculum: What 1-Day and 2-Day Cover
