THE OPEN WORKSHOP / BY ZACH CHESTER
Free. No signup.Learn AI app building by making one useful thing. Trace the code, break the workflow, and prove your fix.
Six guided lessons. A complete Node app.
No AI key needed for local practice.
“We keep losing enquiries between email and a spreadsheet.”
A teaching preview. Run the source kit to use the real local app.
01 / THE GUIDED BUILD
Build an AI enquiry assistant: a real form, persistent records, optional drafts and a protected operator inbox.
Bring a code editor, a terminal and Node.js 22.22+. You will read and change JavaScript files. New to these tools? Start with MDN’s environment setup, then return to lesson 01.
The source app runs without packages or an AI key. Optional live AI uses your own provider account and can incur charges.
Saved only in this browser. No account. These are your self-checks, not a qualification.
Trace a request from a form to a saved record and a human review.
Read → change → check ↗Keep the original request when an optional AI step fails.
Read → change → check ↗Distinguish valid output structure from a factually supported draft.
Read → change → check ↗Check permissions on the server and separate reviewed from sent.
Read → change → check ↗Test a failure without losing the record or calling a paid provider.
Read → change → check ↗Verify storage and the complete journey on the deployed system.
Read → change → check ↗02 / THINK IT THROUGH. THEN TRY IT.
Choose a design. Run a local simulation. Read the trace to see what your choice changes. No setup, model calls or real customer data.
A fictional request arrives. The optional draft step times out. Choose the order of operations before running it.
Keep the original before attempting optional work. In the full source app, an operator requests the draft separately after the enquiry has been saved.
await draft();
await save();
// or
await save();
await draft();A private inbox contains one fictional enquiry. Choose the access boundary to test.
A hidden link changes navigation. A server permission check controls whether the records leave the server.
hideAdminLink();
return enquiries;
// or
if (!session) return 401;
return enquiries;Original: “We need an enquiry inbox. Timing and budget are undecided.” Fixture draft: “Build an inbox for $200.”
A schema can accept the right fields with the wrong facts. Keep budget and timing unknown until the sender establishes them.
validateShape(draft);
accept(draft);
// or
validateShape(draft);
compareWithOriginal(draft);03 / ALREADY BUILDING? START WITH THE FAILURE.
Find the source of truth and prove persistence with a restart and concurrent-write test.
REPAIR 02 ↗Use a direct request to verify server-side access control for an operator inbox.
REPAIR 03 ↗Prevent a polished AI summary from turning absent dates, budgets, or commitments into facts.
01 / FEATURED TOOL · FREE & OPEN SOURCE
Give your coding agent Shopify release context alongside the docs. Search official updates and find candidate changes relevant to your project.
shopify-updates check /path/to/project --task "add order tracking"The check points to matching project files and release entries. Read the evidence before changing code.
v0.2.0 release notes ↗BEFORE YOU START
The six guided lessons, browser experiments and MIT-licensed source are free, with no signup. The emphasis is understanding and testing an AI-assisted app, rather than collecting prompts. Optional live AI requests use your own provider account.
You will run a real local enquiry app with saved records and operator review. It works with rules first; a separate lesson explains optional AI drafts. Browser simulations are small teaching models, not the full app.
No. Progress records your own concept and hands-on checks in this browser. The deployment lesson explains the additional storage, access and failure checks needed on an actual hosted pilot.