Fable jailbreak, PRD's are dead, Does anyone know what a Skill actually is suppose to be? | Ep 18
TL;DR
The Fable 5 jailbreak was embarrassingly simple: Ask it to fix your code instead of find vulnerabilities, and it would comply, raising questions about whether pulling the model was warranted.
PRDs are being replaced by prototypes: A problem statement plus a working prototype plus verification criteria is becoming the new standard, letting PMs get customer feedback faster.
Verification is the competitive moat: How you verify outputs in runtimes, code, and various states determines output quality and is what separates good AI engineering from bad.
Terminology is completely in flux: Skills could be thin capabilities or end-to-end workflows, harnesses could be generic orchestrators or focused workflows, and no one agrees on definitions.
Sometimes UI still beats agents: Well-designed interfaces like the Starbucks app can outperform AI agents for speed, and Airbnb actually pulled an agent booking feature because it was too slow and confusing.
More code faster isn't always better: Faster development cycles risk frustrating customers with constant UI changes, and large commits introduce more risk into systems.
The Breakdown
The Fable 5 jailbreak that got the model pulled by the government was absurdly simple: instead of asking it to find security vulnerabilities, you just said fix my code and it would find them as part of the repair. The hosts debate whether PRDs are dead, unpack the confusion around what skills and harnesses actually are, and argue that verification is the real moat in AI engineering.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
The Cheapest Model That Passes
OpenRouter lists 400 models behind one API. The fix for choosing isn't a better leaderboard, it's a four-step protocol that ends in a real eval.

Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.

Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.