The work loop has changed: delegate, verify, iterate
Benchmarks let you delegate larger tasks. They don’t eliminate any checks. Everything here exists to shorten the cycle between asking and proving.
Result, where, proof. Proof comes before the summary, always. Without it, the agent “thinks it’s done.”
A clean Git tree before delegating turns “it got worse” into git checkout and a second round with a new criterion.
A frontier model creates the plan; a cheaper model handles the volume. Same subscription, much more work per week.
What you'll know how to do
Write verifiable task contracts
Result, where, and proof of completion. The template every prompt in the course follows.
Enable Astra across all three surfaces
ChatGPT app, Codex app, and Codex CLI, with config.toml, sandbox, and AGENTS.md verified.
Use computer use without publishing junk
Linked extension, mandatory draft before any public action, visual proof at the end.
Iterate on a real project with rollback
Switch models, refactor the UI with acceptance criteria, revert in one command, and do it better.
Write tool and taste skills
Remotion in one prompt; and a visual guide with numeric rules that makes “I don’t like it” verifiable.
Connect MCP and run long-running tasks
Blender via MCP, with milestones, evidence at each step, and the ability to resume without losing progress.
Manage quotas and orchestrate models
Track the usage rate, choose the right effort level, and send the bulk of the work to the cheaper model.
Give your agents shared memory
A markdown base that every agent reads before making a decision and updates afterward.
Two tracks: operate, then scale
Each module has six topics, diagrams, command blocks, and prompts ready to copy, with “how to verify” instructions, plus a quick three-question quiz.
🧭 Setup and operation
What Astra changes in the work loop, how to activate it across three surfaces (app, Codex, CLI), computer use in the browser with drafts and verification, and the code/iterate/rollback cycle on a real project.
🛠️ Build and scale
Skills (Remotion and a taste library), MCP for long-running tasks (Blender), quota, cost, and orchestration with cheaper models, and shared memory across agents with Obsidian.
For people already running an agent in the terminal who want to stop improvising
For people who use Codex or Claude Code
and you want a contract, sandbox, and verification instead of trial and error for every task.
Developer
that delegates entire tasks in a repository and needs a small diff, acceptance criteria, and rollback.
For people who write skills and connect MCP
and you want the standard for descriptions, third-party audits, and long-running tasks with milestones.
For whoever pays the bill
and you need to plan for quota, choose the right effort level, and orchestrate cheaper models for volume.
More than reading: a learning platform
Everything runs in your browser, with no login. The course remembers where you left off.
Automatic progress
Mark topics as read and track your progress by module, track, and course.
Notes and highlights
Select any passage to highlight or annotate. It’s saved in your browser.
My journey
Pick up where you left off and see everything you’ve studied in one dashboard.
Themes and reading
Dark, light, sepia, focus, high contrast. Font size and line spacing.
Quiz by module
Three questions with answers and explanations to check your understanding.
Prompts with proof
Each workflow includes the translated prompt, the placeholders to replace, and what to check at the end.
Where the claims in this course come from
Model launch
Official GPT-6 Astra page (OpenAI): agent capabilities, benchmarks, and cost per task completed. The launch figures appear in module 1.1, along with what they don’t measure.
Commands and configuration
Codex documentation. Every command in this course was verified in Codex CLI 0.153 in September 2026. Flags and keys change between versions: codex --help on your end is the final source.
Open-source projects used in the examples
Remotion (video in React), Blender MCP (3D modeling via protocol) and Obsidian (markdown notes base).
Limits
Benchmarks don’t measure your project. Menu names change with each app version. Quota usage depends on your mix of tasks, so measure it instead of relying on example figures. And no prompt replaces reading the draft before publishing.
Ready to delegate the entire task and verify the result instead of hoping for the best?
Start with Track 1. In about 5.6 hours, you’ll come away with the task contract, Astra enabled across all three surfaces, the iteration and rollback cycle, and a complete system of skills, MCP, orchestration, and memory.