06-reference

every tale two models

2026-07-05·reference·source: Every·by Kate Lee

"A Tale of Two Models" — @Kate Lee

Why this is in the vault

Every's Context Window team ran live side-by-side usage of Fable 5 and Sonnet 5 in the week both launched. The finding is directly RDCO-relevant: Fable 5 immediately justified itself (the team rebuilt their document editor Proof in ~3 hours from a single prompt), while Sonnet 5 failed to carve out a defensible role against Opus 4.8, Fable 5, and GPT-5.5. The Alignment section by Ashwin Sharma surfaces Anthropic's drug discovery pivot — Claude Science plus internal preclinical programs — as a "sell the shovels" platform strategy, which is a meaningful signal on how Anthropic is evolving beyond a pure API vendor.

⚠️ Sponsorship

No paid third-party sponsor in this issue. Heavy self-promotional CTAs for Every-affiliated products appear throughout: Spiral (AI writing), Monologue (voice dictation), Sparkle (file organizer), Cora (AI email client), Proof (document editor), Every Studio umbrella, Q2 Demo Day (July 10, paid subscribers), and Every IRL Brooklyn meetup (July 15). All coverage of these products should be read as house-promotional.

Issue contents

Context Window lede — "A Tale of Two Models" (Kate Lee)

Comparative assessment of Fable 5 vs. Sonnet 5 from the Every team's actual usage. Fable 5 returned from its government-enforced suspension and immediately proved its value. Sonnet 5, positioned as the everyday workhorse, found no clear win role — priced above GPT-4.1, slower than Fable 5, less capable than Opus 4.8. Katie Parrott's "Vibe Check" section reinforces: no task where Sonnet 5 is the obvious choice given the current competitive field.

AI and PowerPoint — Mike Taylor

Built a 24-step pipeline to automate consulting deck production at $62/deck. Conclusion: technically achievable, not broadly recommendable — complexity and cost floor make it viable only for high-volume, high-value contexts.

Codex in Practice — Dan Shipper, Austin Tedesco, Kieran Klaassen, Natalia Quintero

Four Every team members describe their distinct Codex workspaces. The standout case: Natalia Quintero uses Codex as a direct report for inbox management, client pipeline tracking, and medical care coordination — a nontechnical builder extracting agentic leverage without writing code.

AI Strategy as Explicit Betting — Dan Pupius

Framework: four variables to actively manage — token costs, model capability, provider lock-in risk, and regulatory exposure. Treat model selection as a portfolio bet with conscious position sizing, not a permanent infrastructure decision.

Alignment — Anthropic and Drug Discovery (Ashwin Sharma)

Anthropic launched Claude Science (desktop research tool for scientists) and announced internal preclinical drug programs. The goal is not to compete with pharma but to dogfood Claude on hard, slow biological verification problems — building the workflow layer (data, models, experiments, decision loops) and positioning it as a platform to license to big pharma. Classic "sell the shovels" strategy.

Mapping against Ray Data Co

Fable 5 vs. Sonnet 5 directly informs RDCO agent stack decisions. Every's finding — Sonnet 5 has no clear win role against the current field — maps onto the active RDCO stack question post-launch. For high-complexity agentic work (Claude Code, COO-agent orchestration), Fable 5 or Opus 4.8 remain the call; Sonnet 5 looks viable only for high-volume, latency-sensitive, cost-sensitive batch tasks.

Natalia Quintero's Codex-as-direct-report pattern is a close analog to the RDCO COO agent model — nontechnical coordination tasks (inbox, pipeline, scheduling) running on agentic infrastructure. Her use case is worth tracking as a template for extending agent scope without eng overhead.

Dan Pupius's four-variable betting framework (token costs / model capability / provider lock-in / regulation) is a usable structure for formalizing RDCO's model selection reasoning, especially for phData DSA client advisory work where RDCO helps clients make AI stack decisions.

Anthropic's drug discovery pivot reads primarily as a signal on Anthropic's vertical ambitions beyond the API layer — building proprietary workflow infrastructure — rather than a near-term RDCO investment thesis. Relevant to how RDCO models infrastructure dependency on Anthropic long-term.

Related