One Course
Every lesson, project and mock exam in Coding with AI agents.
Billed monthly. Cancel any time.
Level 1: Newcomer
0 points
Sign in to save your progress across devices.
Courses / AI tools and development
Practice coding with AI agents by deciding what they may change and how you will check the result. Write repository instructions, select context, inspect proposed changes and reject unsafe actions in a scripted browser workspace.
Unit 1, Task contracts: 5 lessons and 26 checked exercises, with no account or card.
Start the free unitThe rest of the course opens with a One Course or All Access plan. See plans and prices
Choose access here. Your course is already selected.
Complete the incomplete refund brief with bounded instructions, grounded evidence and preserved task scope.
6 steps · 12 minCase study · BeginnerThe leaked configuration caseComplete the leaked configuration case with bounded instructions, grounded evidence and preserved task scope.
6 steps · 12 minProject · IntermediateA portable repository handoffComplete project: a portable repository handoff with bounded instructions, grounded evidence and preserved task scope.
8 steps · 18 minCase study · IntermediateThe poisoned skill caseComplete the poisoned skill case with bounded instructions, grounded evidence and preserved task scope.
6 steps · 12 minProject · IntermediateA safe balance fixComplete project: a safe balance fix with bounded instructions, grounded evidence and preserved task scope.
8 steps · 18 minProject · AdvancedA grounded review queueComplete project: a grounded review queue with bounded instructions, grounded evidence and preserved task scope.
8 steps · 18 minCase study · AdvancedThe parallel ownership clashComplete the parallel ownership clash with bounded instructions, grounded evidence and preserved task scope.
6 steps · 12 minProject · ExpertA bounded delivery runComplete project: a bounded delivery run with bounded instructions, grounded evidence and preserved task scope.
8 steps · 18 minMock exam · ExpertTimed job screen: a bounded featureComplete timed job screen: a bounded feature with bounded instructions, grounded evidence and preserved task scope.
12 questions · 35 minMock exam · ExpertTimed job screen: unsafe-run recoveryComplete timed job screen: unsafe-run recovery with bounded instructions, grounded evidence and preserved task scope.
12 questions · 35 minPractice agent-assisted software work in a deterministic browser workbench. Write task briefs and repository instruction files, scope Codex, Claude Code, Copilot and Cursor workflows, and distinguish skills from enforced gates. Review seeded behavioral bugs, unsafe SQL and weakened tests. Keep secrets out of context, use narrow MCP capabilities, isolate parallel work and bound token usage and retries. Carried projects, original cases, interview sets and two timed job screens check your actual artifacts without live model calls or external services.
Explain what you've learned in a full-screen interview. Speak or type, then review your answers against the rubric for your level.
Interview loops by jobBeginner
Bound a request, gather repository context and define evidence before delegating.
Write scoped durable guidance and distinguish the supported instruction mechanisms.
Keep secrets out of context and separate task authority from hostile repository text.
Intermediate
Apply dated Codex, Copilot and Cursor instruction mechanisms without treating them as identical.
Package reusable procedures, load relevant resources and enforce risky write boundaries.
Keep patches narrow and preserve tests while responding to grounded review findings.
Advanced
Find behavioral defects, security regressions and weakened tests in seeded proposals.
Use narrow MCP capabilities and handle tool output without granting it authority.
Assign worktrees and file ownership, then verify the combined change.
Expert
Bound usage and retries, then leave a compact reproducible handoff.
Check dependency provenance, approved data use and incident evidence before release.
Compare prompt revisions on held-out cases and maintain an auditable operating procedure.
A useful task names the behavior that should change, the files in scope and the evidence you expect back. An instruction such as “make it better” leaves all three decisions to the agent. For a fictional returns service, ask for a repair to duplicate return requests, require the existing success response to stay the same and name a test that sends the same request twice.
Use the AGENTS.md examples to practice repository guidance. The Claude Code tutorial covers CLAUDE.md, while the Codex tutorial follows a bounded change through its handoff. Product conventions were checked on November 14, 2026; verify the installed version before applying them at work.
A reusable skill needs a clear trigger and a procedure you can inspect. The skills guide distinguishes a skill from repository instructions. Practice code review by checking a proposed diff against the acceptance condition, including the case that failed before the change.
A passing test can still miss the defect. Look for a weakened assertion, a skipped test or a mock that hides the behavior under review. The exercises include incorrect assistant replies so you have to challenge the evidence.
The course includes GitHub Copilot instructions and Cursor rules. The Claude Code and Cursor comparison explains which conventions belong to each product. Practice the underlying decisions here; use the vendor’s environment for installation, model settings and live execution.
Start with the coding-agent exercises. The work stays fictional, and the assistant’s scripted response does not predict what a live model will produce.
Use the Claude Code hooks guide for event checks and the worktrees guide for parallel file ownership.
No. You work in our own scripted practice workspace. Product-specific lessons explain instruction conventions, but the course does not install vendor software or call live models.
The replies are scripted for each exercise. You can inspect the scenario and practice a decision, but this is not an open chat or a benchmark of a vendor model.
You need enough programming knowledge to inspect a change and explain a test. Use the programming courses alongside these agent tasks.
No. The GitHub Copilot certification guide separates our independent practice from the official credential.