Agent Simulator
Practice designing AI agents the honest way: write an agent’s loop for a real scenario, or walk the loop turn by turn picking its next move. Get a deterministic 0–10 score with per-dimension feedback, and see side-by-side what sloppy vs. disciplined agent design earns. Companion to the Advanced Prompt Engineering video course.
Before you start
Teaching simulation — no AI runs here. The agent loops, tool outputs, and observations are hand-crafted examples showing what sloppy vs. disciplined agent design earns, so every student sees the same lesson. Nothing you type leaves your browser. A real-agent playground is a separate future project.
Choose a practice scenario
Scenarios unlock in order across Episodes 1–8 — score 4+ to open the next one. Your best scores are saved on this device.
Scenario
Tip: break the task into ordered steps, name a tool per step, say what happens on failure, keep instructions separate from data, and define each step’s handoff.
Scenario
Your score
New personal best — saved on this device.
Feedback per dimension
See the contrast
Both runs below are hand-written teaching examples — the same ones every student sees. No AI generated them.
What a sloppy agent loop does
Typical outcome
What a disciplined loop does
Typical outcome
Exemplar
You built a disciplined agent.
You scored 8/10 on the capstone — that unlocks your course certificate.
Opens the Certificate Maker with “Advanced Prompt Engineering” pre-filled — add your name, then print.
How to practice agent design here
- Pick a scenario — start with “The Wall” from Episode 1. Later scenarios unlock as you score 4+.
- Design scenarios: write the agent’s loop — ordered steps, a tool per step, failure handling, guardrails, handoffs. Walk-the-loop scenarios: pick the agent’s next move each turn; wrong picks explain why.
- Score it — a deterministic 0–10 with one line of feedback per dimension. The same input always earns the same score.
- See the contrast to compare hand-written examples of a sloppy loop vs. a disciplined loop.
- Reveal the exemplar, study the highlighted scoring cues, and retry to beat your best — best scores are saved on your device.
- Capstone: score 8+ on Episode 8 to unlock your course certificate.
Teaching simulation — no AI runs here. The agent loops, tool outputs, and observations are hand-crafted examples showing what sloppy vs. disciplined agent design earns, so every student sees the same lesson. Nothing you type leaves your browser. A real-agent playground is a separate future project.
Frequently asked questions
Is the Agent Simulator a real AI agent?
No. It is a teaching simulation — no AI runs here. The agent loops, tool outputs, and observations are hand-crafted examples showing what sloppy vs. disciplined agent design earns, so every student sees the same lesson. Scoring uses deterministic keyword checks and fixed answer keys, not a model.
Is what I type private?
Yes. Nothing you type leaves your browser — there are no network requests, no accounts, and no servers. Only your best score per scenario is saved, in your own browser’s local storage.
How is my agent design scored?
Design scenarios check five dimensions — Planning, Tools, Recovery, Guardrails, and Output — for scenario-specific keywords, awarding 0–2 points each for a total out of 10. The same design always earns the same score.
How do the “walk the loop” scenarios work?
Each turn shows the agent’s current state and several possible moves. Pick the right move to earn that turn’s points; wrong picks explain why the move fails. Points per turn add up to 10.
How do I unlock new scenarios?
Scenarios unlock in order across Episodes 1–8: score 4 or more out of 10 on the current scenario to unlock the next one. Your best score per scenario is remembered between visits.
How do I earn the certificate?
Score 8 or more out of 10 on the Episode 8 capstone, “Capstone: Build the Agent”. A “Claim your certificate” button then links you to the site’s Certificate Maker with the course name “Advanced Prompt Engineering” pre-filled.
More from the codex
Agent Simulator
Practice AI agent design with deterministic scoring.
Prompt Simulator
Practice prompt engineering: write prompts, get scored 0–10.
Open tool → LiveCertificate Maker
Make a printable award certificate in seconds.
Open tool → LiveFlashcard Maker
Spaced-repetition flashcards for any subject.
Open tool →Support the codex
This tool is free, with no account and no ads. If it saved you some time, consider supporting the indie dev behind it.
Questions or a bug to report? Email [email protected].