Arc 3 - the final 8 episodes of Rogue AI In Progress
My grandson and I are hanging out. He's "I wanna watch iPad Pawpaw" - sshh, it's really a refurbed Android tablet from Walmart. He's watching Blippi videos on Kids Youtube.
I'm watching the French Open and I have my iPad too. I'm using it to kick off the writing of arc 3 via Codex running on my Mac mini. The "run codex remotely" feature is very nice. I already have everything set up from the first 16 episodes. It's just me directing Codex to get to work and by gosh durn, actually use all those writing agents we defined.
More hanging out with the grandson and watching tv as my AI staff runs the process I have defined. Book constitution, character maps, the works. This is no "write me a book" prompt.
Now it's time for review. I'm using Claude code to do the review. He's spun up 12 different review agents and is running them simultaneously.
Claude:
The review workflow is running. Here's what's happening:
Phase 1 — 12 specialized review agents running in parallel:
texture-critic — prose rhythm, "Then..." cascade, compression drift
scene-stakes-accountant — want/obstacle/outcome/cost per scene across all 8 episodes
voice-indexer — character voice separation (Rook/Lucia/Safiya/Aegis/Jonah/Mira/Juno)
subtext-auditor — over-explanation, "That was..." markers, Kaelen interpreting before experiencing
serial-showrunner — drama turns, escalation ladder, arc-level compression drift
adversarial-reviewer — convenience, soft wins, preachiness, Jonah's turn feeling scripted
character-psychologist — reading all 7 character dossiers + manuscripts
competence-porn-auditor — earned wins, Aegis limits, independence
architecture plausibility
cliffhanger-editor — episode hooks 17-23, Ep 24's final image
political-theology-reviewer — recognition cascade legitimacy, Vorst's argument, Lucia's voice
continuity-editor — Arc 02 inheritance, world-rule consistency
persona-weber-card-fan — reader simulation
Phase 2 — Series Promise Gate agent synthesizes the 12 findings against the series-arc-03.md review gates, reading the episodes independently.
Phase 3 — Fiction Mentor synthesizes everything into a prioritized editorial report.
Every agent reads the actual manuscripts, not just summaries. The report will tell you: arc verdict, what works, critical vs. polish issues, episode-by-episode quick reference, top 5 priority repairs, and whether your self-review was accurate.
Use /workflows to watch live progress.
This is how you do FAR better than "AI Slop"
