“Model Wars” — @Laura Entis (Every / Context Window)

Why this is in the vault

Three discrete extractables in one issue: (1) GPT 5.5 launch reframed as a product question, not just a model question — useful framing for the RDCO position that AI bets compete on harness/distribution, not raw capability; (2) a quantified “AI tell” (the “not just a ___, it’s a ___” construction up 4x in US corporate docs 2023-2025) — a concrete editing rule for Sanity Check drafts and the voice-match skill; (3) Austin Tedesco’s Remotion-based product-video workflow — a candidate workflow for RDCO’s own product clips. Body did not render via Gmail; reconstructed from canonical URL — flag for manual review on first pass.

The core argument (lead essay)

OpenAI’s GPT 5.5 launch is a competitive milestone, but the OpenAI-vs-Anthropic battle is “a product question as much as a model question.” Speed, collaboration affordances, and engineering quality matter — not raw benchmark wins. Implicitly: Anthropic’s Claude Code / Cowork product surface is the moat, not the model weights themselves.

Issue contents

Inside Every — “Codex versus Claude Code,” debate between Nityesh Agarwal and Naveen Naidu
Data point — “not just a ___, it’s a ___” construction up 4x in US corporate documents 2023-2025; humans now actively avoid it to dodge AI-tell accusations
Steal this workflow — Austin Tedesco’s product video pipeline: screen-record usage → Claude/Opus drafts storyboard → coding agent builds in Remotion → first render usually shippable
Now, next, nixed — “Prompts are the new installers” framework on software install trends
Model happenings — Cowork live artifacts, Codex screen memory, OpenAI workspace agents
One last thing — ChatGPT Images 2.0 creations by Nityesh Agarwal

Mapping against Ray Data Co

Strong — voice/editing rule. The “not just a ___, it’s a ___” construction is now a measurable AI tell. This is a concrete addition to the founder’s voice-match and draft-review skills: flag the construction in Sanity Check drafts and rewrite. Pairs with the existing pattern of avoiding em-dashes-as-AI-tells; the bar keeps rising on what reads as machine-written. See ~/.claude/skills/voice-match/SKILL.md and ~/.claude/skills/draft-review/SKILL.md — both warrant a one-line addition for this construction.

Medium — product video workflow. Remotion + storyboard-via-LLM + coding-agent-builds-render is exactly the kind of pipeline RDCO would use for product demo clips (Sanity Check teasers, HQ explainers, RDCO landing-page hero). Worth queuing as an investigation: does the same pipeline work for marketing micro-videos at sub-1-min length, and what’s the iteration cost vs After Effects / Descript?

Medium — model-vs-product framing. Reinforces RDCO’s harness-thesis: model capability commoditizes, the wrapper / workflow / distribution surface is where edge persists. Aligns with Thariq’s Apr 15 Anthropic guidance on context engineering and with the “thin harness, fat skills” position.

Skip — Now/next/nixed prompts-as-installers. Cute framing, no actionable hook for current bets.

Curation section — notes

Inside Every — Codex vs Claude Code debate. Self-cross-promo (both authors are Every contributors; Naveen Naidu and Nityesh Agarwal both write under the Every umbrella). Not third-party curation — flag as internal cross-promo.
Steal this workflow — Austin Tedesco. Tedesco is Every’s Head of Growth; this is internal team write-up, not third-party. Still useful, but disclose: the workflow is Every’s own pipeline, not externally validated.
Model happenings. Mix of Every-owned product news (Cowork artifacts) and external industry items (Codex, OpenAI workspace agents). The Cowork item is owned-product promo; flag accordingly.
One last thing — Nityesh Agarwal images. Internal contributor showcase, not external curation.

Owned-product mentions across the issue: Proof, Sparkle, Monologue, Cowork — all carry utm_source=every&utm_medium=banner&utm_campaign=post. No explicit third-party sponsor disclosure block appeared.

2026-01-22-every-cursor-future-of-code — prior Every framing on coding-agent product surface
2026-02-02-every-codex-vs-claude-code — earlier Codex/Claude Code comparison from the same publication
2026-01-27-every-compounding-engineering-intro — Every’s “compounding engineering” framing pairs with the model-as-product thesis
~/.claude/skills/voice-match/SKILL.md — candidate update site for the “not just a ___, it’s a ___” AI-tell rule
~/.claude/skills/draft-review/SKILL.md — same