Speed and trust
Humans own the design and verification. Agents carry implementation. Specs, tests, review, observability, and rollback make speed durable.
A field guide to AI coding anti-patterns
AI can ship code at absurd speed.
The mess arrives just as fast.
Slopapalooza names the failure modes that appear when powerful coding agents meet weak engineering systems — and shows serious teams how to keep the speed without inheriting the wreckage.
Not anti-AI. Anti-unforced-error.
19named failure modes
90 secteam risk diagnostic
9production-readiness checks
The point, without the costume
This is not an AI backlash.
It is a higher bar.
AI coding is extraordinary. That is exactly why the operating model around it matters.
The danger is not one weird generated function. It is the system around it: vague intent, thin review, broad permissions, no rollback, and nobody who can explain what just shipped. Each shortcut looks harmless. Together, they compound.
It looks fine on day one. The invoice arrives later, with interest.
Field guide / selected entries
The mess has a pattern.
Once you can name a failure mode, you can design a system that prevents it.
01
Planning
Prompt & Pray
Vague intent goes in. Confident code comes out. Nobody checks whether the architecture makes sense.
Read the pattern → 02Architecture
The 9,000-Line Component
The agent keeps adding to the file that already exists because nobody stopped to redesign the thing.
Read the pattern → 03Review
The Rubber-Stamp Stage
A 1,200-line pull request, a three-bullet AI summary, and the most expensive possible interpretation of “LGTM.”
Read the pattern → 04Operations
Autopilot With No Pilot
An agent takes irreversible customer-facing actions with no approval gate and no kill switch.
Read the pattern → 05Verification
Coverage Theater
The AI wrote the code and the tests. They agree with each other. Unfortunately, they are both wrong.
Read the pattern →The complete catalog
Fourteen more ways to accidentally build tomorrow’s incident.
Browse all 19 anti-patterns →The antidote
Move human judgment up the stack.
Use AI for more implementation, not less. Put people where judgment matters most: intent, architecture, constraints, and proof.
Use the 9-point pre-flight checklist- 01Specify the outcome
Write down behavior, constraints, failure states, and acceptance criteria before generation begins.
- 02Constrain the system
Give agents the context and permissions they need — not the keys to every kingdom.
- 03Verify the behavior
Test the result against reality, not against the same assumptions that produced the code.
- 04Keep a human owner
Someone must be able to explain the change, operate it, and reverse it when the vibes deteriorate.
The 90-second diagnostic
How close is your team to Slopapalooza?
Seven direct questions about specs, review, tests, ownership, and production controls. No email gate. No twelve-page PDF at the end.
Take the Slop Risk Test →Field reports
Seen some slop?
Inherited an API designed entirely by autocomplete? Survived a migration with no rollback plan? Tell us. The best worst stories get published, anonymized on request.
Submit a horror storySame thesis, different costume