Home › Courses › Build an Autonomous Coding Agent
Build an Autonomous Coding Agent
Your own Claude Code: plan, edit, test and commit, built from scratch
1 phase. 6 lessons. 6 labs. Build an advanced CLI coding agent from raw API calls: an AST-based file relevance scorer, a validated planning engine, safe backup-verify-replace file modification, a test-driven fix loop, multi-file rollback, and git integration — then run it on a real task in your own repo.
6 lessons · 6 labs · Advanced · 1 real coding agent
Outcomes you will have by the end
- A real coding agent that ships a real commit — Point it at your own repo, give it a real task, and watch it plan, safely edit multiple files, run your real test suite, iterate on failures, and commit — with a summary you can actually verify, not just trust.
- A working mental model of how Claude Code and Devin work — Codebase relevance scoring, structured plan validation, backup-verify-replace editing, and test-driven fix loops are the real architecture behind every serious coding agent on the market.
- Reusable safety patterns for any autonomous agent — Backup-before-write, all-or-nothing multi-step rollback, and hard-capped retry loops are patterns that apply to any agent that takes real, hard-to-reverse actions — not just coding agents.
- Verified certificate of completion — Issued by DeVenture Academy, tied to your completion record. Lists the AST-parsing, structured-planning, and safe-execution skills you built.
What you will be able to do
AST-Based Code Understanding · Structured Plan Validation · Safe File Modification · Test-Driven Fix Loops · Multi-File Rollback · Git Integration
Every phase, every lesson, every project
- Build an Autonomous Coding Agent (6 lessons) — Coding agent architecture, AST-based codebase understanding, structured planning with validation, safe execution and test-driven fix loops, multi-file rollback, git integration
The technologies you will use
Anthropic Claude API · tree-sitter · Pydantic · pytest · GitPython
Roles this course prepares you for
- Senior Developer / Tooling Builder (Varies by employer) — Build a customized coding agent tuned to your own team's codebase conventions, test setup, and workflow — beyond what off-the-shelf tools offer.
- AI Tooling Consultant ($100-250/hr) — Help engineering teams design and safely deploy their own internal coding agents, with the safety patterns (backup, rollback, validation) this course builds explicitly.
- Open-Source Contributor (N/A — reputation and portfolio value) — Extend or fork this agent pattern for a specific ecosystem (a different language, a different test framework) and publish it.
Common questions
Do I need to complete Course 01 first?
Strongly recommended. This course assumes you already understand the basic tool-use/agent-loop pattern from Build Your First AI Agent from Scratch and moves quickly into codebase-specific complexity on top of it.
How much does this cost to run?
Budget for it: a single realistic task through the finished agent typically costs 50k-200k tokens, roughly $0.50-$2.00 at Sonnet pricing — and you'll run it more than once while debugging. This is flagged explicitly in Lesson 1, not buried in the fine print.
Is this safe to run on my real projects?
The course is built specifically around the safety patterns that make this reasonable: backups before every write, full rollback on any failure, and plan validation before execution. Still, start on a repo you have backed up elsewhere or that's under version control with a clean working tree.
What do I end up with?
A real CLI tool, `code_agent.py "<task>"`, that reads your repo, plans a change, safely edits the relevant files, runs your real tests, iterates on failures up to 3 times, and commits with a generated message and a verifiable summary of what changed.
Key terms in this course
Agent
Continue your learning path
Agentic AI Engineering
Start the Build an Autonomous Coding Agent course
Create a free account — the opening phases of 24 of 30 courses are free, no credit card. Or see Pro pricing.
All courses · Pricing · About · FAQ · Glossary