Every skull was a JS framework.
Modern Coding School teaches you to judge, direct, and ship AI-built software to a production standard — by reading the autopsies of everything that shipped without judgment. People's servers died for your knowledge.
Self-hosted. Sovereign. No platform owns your credential, and no model you can't defend ships to prod.
Technical judgment, without
technical production.
Read, don’t write
Control flow, data flow, diffs, stack traces. Understand the code you didn’t type — and the autopsy of the one that died.
Specify
Turn a vague want into executable agent instructions a machine cannot misread.
Verify
Trust output you didn’t write — and prove it clears the bar before it ships anywhere near prod.
One core. Two tracks.
Every exit is defended.
Every concept is published free, forever. We sell the with-you: the feedback, the lab, the gate, the credential you actually own.
Technical Judgment
The shared foundation: reading code you didn't write, specifying work for agents, verifying output, and clearing a production bar.
- Read a diff and a stack trace
- Spec work an agent can execute
- Verify output against a source
- Clear the production bar
Production & Self-Host
Deploy and operate a real system on your own metal: domains, TLS, secrets, backups, logs, rollback.
- Own the deploy and the fire
- Logs · monitoring · rollback
- Secrets and TLS without a SaaS
- Exam: find and fix a real bug
Agent Company
Orchestrate agents behind verification gates with real unit economics — and no platform or model lock-in.
- Verification gates that actually gate
- Token cost and model routing
- Self-hosted defensible models
- Exam: a business that ran 30 days
An agent-operated school that
doesn’t depend on a platform.
Software does the first pass of everything; one human owns the judgment. It runs self-hosted, on open models we can defend. The business is the demo — you're reading it.
Incident Days
We inject real failures into your workspace — a leaked secret, a runaway agent loop. You diagnose under time pressure.
Rubric-as-code
The gate: automated checks + agent first-review + human sign-off. First-pass feedback in minutes, not days.
The signed credential
Your diploma is a cryptographically signed artifact on your own domain — verifiable, federated, owned by you.
Run on your metal
The whole stack is FOSS and self-hostable. Learn to operate it without renting anyone's cloud or trusting anyone's model blindly.
Learn on other
people’s mistakes.
This AI-built app leaked every user's email. Here's the line.
A citizen-built CRM. One missing auth check. 14,000 records walked out the front door.
Read the autopsy →Five questions to ask before you trust your agent's output.
A reading list for the diff you didn't write.
The runaway agent that spent €4,200 in a weekend.
No spend cap, no kill switch, a recursive tool loop.
The migration that dropped the table it meant to rename.
An agent-written migration. A staging that wasn't prod-shaped. A Monday.
The login that let everyone in if you left the password blank.
A truthy check on an empty string. Generated, approved, deployed.
One autopsy, one judgment lesson, every week. Subscribe by e-mail or RSS — we won’t build a moat out of your inbox.
A credential no
platform can revoke.
Every entry is signed, self-hosted on the graduate’s own domain, and federated. Verify it from the command line. Click one — it runs.
--BEGIN MCS CREDENTIAL-- holder: Dana Okafor degree: M. Agent Operations track: B · cohort MMXXVI-1 capstone: ledgerloop.app [200 OK] gate: CONFERRED · score 94 key: 9F2A 4C71 … E0B3 --END MCS CREDENTIAL--
Don’t add to
the pile.
The application is the first exercise: spec a small build with any agent, submit the result — and your own evaluation of it. The admissions agent scores your judgment; a human makes the call.

Questions we get asked
Updated as new ones come in. Email us if yours isn't here.
Who is this for?
What does the gate mean?
Is the content free?
What is a signed credential?
What does self-hosted mean here?
When does the next cohort start?
Don't add to the pile.
The application is the first exercise. Spec a small build with any agent, submit the result, and your own evaluation of it.