The definitive text-first course on Claude Code, Cursor, and Google Antigravity. Terminal access, efficient search, parallel agents via git worktrees, stopping conditions, and building your own agent loop. With links to every open-source implementation and reference video worth your time.
This is a text-first course that links out to the best supporting material on the internet instead of trying to replace it. The goal is to make this the best course on agentic coding IDEs you can find — even without producing a single minute of custom video.
This course is built by people who use Claude Code, Cursor, and custom agent loops every day to ship production AI systems. It reflects how these tools actually behave on real codebases — not how the landing pages describe them.
Every day has working code snippets you paste into a terminal and run right now. You will spin up a git worktree, run ripgrep against your own codebase, and build a toy agent loop before the week is over.
Instead of shooting videos we would have to re-shoot every six months, this course links to the definitive open-source implementations (Cline, aider, OpenHands, SWE-agent) and the best conference talks and deep-dive videos on the topic.
Each day is designed to finish in about an hour of focused reading plus hands-on work. You can do the whole course over a week of lunch breaks. No calendar commitment, no live classes, no quizzes.
Each day stands alone. Read them in order for the full picture, or jump straight to the day that answers the question you have today.
The shell tool is the primitive everything else builds on. Learn how Claude Code, Cursor, and Antigravity spawn subprocesses, capture output, gate permissions, and stream results back into model context without blowing the budget.
Why every agentic IDE replaced grep and find with ripgrep and fd. How to structure search output so agents don't burn 40,000 tokens on noise. When to reach for ast-grep for syntax-aware matching that regex can't do.
The 15-year-old git feature that makes parallel agents possible. Learn to spin up N agents in N isolated working directories on the same repo. No merge conflicts, no lock files, no clever coordination protocol needed.
Starting an agent is easy. Stopping one is the actual production problem. The four stopping strategies every serious agent combines: explicit done tool, success check, budget cap, and max iterations.
Capstone day. Fork Cline, strip it to the bare loop, wire in Claude Sonnet 4.6, and ship a minimal agentic coder you understand every line of. ~120 lines of Python. Works against any repo.
Instead of shooting our own videos, we link to the best deep-dives already on YouTube. Watch them alongside the course. All external, all free, all from builders who ship this stuff.
Search results for builder-made walkthroughs of Claude Code's tool use, permission modes, and real-world workflows.
Video tutorials from the Anysphere team and community on how Cursor's multi-file editing and agent mode actually work.
The research team walks through how their "agent-computer interface" differs from a human shell and why that matters for LLM tool design.
Paul Gauthier and community demos of aider's git-integrated edit loop, one of the cleanest implementations of an agent coding workflow in the open.
For anyone who has never used git worktrees — start here. Short, practical, no agent context required. Prerequisite for Day 3.
Deep dives on why ripgrep is 10-50x faster than grep, how its SIMD-accelerated scanning works, and when to reach for it versus git grep.
The best way to understand any agentic IDE is to read one that is fully open. These four repositories implement every pattern in this course. Clone them, trace the code, modify the loop.
Formerly "Claude Dev." The most readable implementation of an agentic coding loop as a VS Code extension. Shell tool, file tools, browser automation. Every tool call visible in the sidebar while the agent runs.
Terminal-based AI pair programmer. Git-integrated (auto-commits), diff-based edits, multi-file context. Small codebase you can read end-to-end in an afternoon. Ideal first source-code study.
Open-source "AI software engineer" platform. Sandboxed execution, browser control, multi-agent coordination. Production-scale agent architecture with isolation, recovery, and state management.
The Princeton research agent that showed LLMs could resolve real GitHub issues. Custom "agent-computer interface" — terminal commands designed for agents, not humans. The canonical paper on why agents need different tools.
You use them. You want to know what's happening under the hood so you can debug when they misbehave and use them more effectively.
You want to ship an agentic product. This course gives you the exact architecture to copy, with pointers to every open-source implementation that proves it works.
You need to pick a tool for your team. This course explains the real differences (not the marketing differences) between the major options.
The 2-day in-person Precision AI Academy bootcamp covers real agent engineering — tools, loops, worktrees, deployment — hands-on with Bo. 5 U.S. cities. $1,490. 40 seats max. June–October 2026 (Thu–Fri).
Reserve Your Seat