Every - Model Has a Six-month Shelf Life Now
Why this is in the vault
Every's own analysts turn one number buried in Anthropic's S-1 — models older than six months earn less than 20% of its revenue — into a live argument about model-switching behavior and the economics of frontier-model dependence, with a secondary read (Thorsten Ball's "you pay for the gap in retries") that lands directly on RDCO's verification-gate thesis.
The core argument
Anthropic filed to go public on Monday, and its S-1 quantifies something Every's staff had only felt anecdotally: when a new model ships, heavy users switch almost immediately. The filing's $16.3B first-half-2026 revenue predates Opus 5.5 and Fable 5.1 — "this is all Opus 5 money," per Every's head of evals Mike Taylor — and customers from a year ago now pay roughly nine times as much. Writer Evan Armstrong (whose full S-1 teardown is already filed separately, see Related) frames this as Anthropic behaving like a movie studio: only as good as its next release. The risk: Anthropic has committed to ~$155B/year in compute by 2028, mostly non-cancelable, while customer spend is not — if model quality gains ever stall ("the discernment horizon," per Steve Yegge, where most tasks aren't hard enough to show a difference between generations), that compute commitment flips from inventory to debt.
Every CEO Dan Shipper extends this into an "inchworm" model of AI adoption: engineers (leading edge) stretch forward into frontier capability first, knowledge workers follow, the rest of the economy catches up last. The piece's practical takeaway, credited to engineer Thorsten Ball: a model that's wrong 10% of the time doesn't tell you which 10% — so you pay for that gap in retries and rework. Every's own prescribed habit matches: read the Vibe Check on every release, and if relevant, switch your default and test it against one hard task you already know well (a "personal benchmark") rather than trusting benchmark marketing.
Curation section
- "The Every Podcast" — Fred Graver on writing with AI: Emmy-winning TV writer Fred Graver (The AI Writer's Studio) argues AI can help structure a story but can't supply a point of view — he briefs different models for different jobs (a "very unpleasant" Claude agent for accountability, ChatGPT as the eager-but-naive film student, NotebookLM as cheerleader) and holds that the human quirks (Homer Simpson's ad-libbed "D'oh!") are what AI still can't originate.
- Skill Share — "TV and Film Pitch Deck Prep": a Claude skill (by Graver) that walks a story idea through a structured pitch-deck build using the Kenn Adams Story Spine, numbering and version-tracking each creative decision (marking superseded choices) rather than silently overwriting them.
- Steal This Workflow — moving a site off Webflow with Claude: Every's Mike Taylor had Claude export his Webflow site, reverse-engineer the design, and migrate the domain to Cloudflare Pages (~$5/mo vs. a $360 renewal) over one morning coffee, including Claude directly driving his browser to move the domain on GoDaddy.
- Inside Every — the "shoggoth strategy": a debate between Mike Taylor (the market pays for a narrow, legible specialty; hide generalist range behind one headline skill) and Dan Shipper (AI is strongest at predictable specialist work and weakest at messy novel problems, so generalists have the real edge) — resolved as "shoggoth mullet": lead with the specialty that gets you in the door, use the generalist range once inside.
- Overheard at Every: Microsoft's buyout offer to ~25% of staff (read as an AI-strategy cleanup, not jobs-for-agents evidence); more field notes from Thorsten Ball's agent-harness post; a one-line "find bad vulnerabilities" prompt matching professional security researchers' bug-hunting results; OpenAI claiming three more AI-generated proofs (math, CS, physics) with outside experts shepherding this round after October's retracted-manuscript pushback.
- Log On — events: a DSPy context-engineering session (Mike Taylor + Drew Breunig + Isaac Miller) and Every's weekly All Access office hours.
- House self-promo (not a paid sponsor): the standard Every footer promoting Every Agent, and the Sparkle/Cora/Spiral/Monologue software bundle — Every's own products, disclosed here as self-promotion, not a third-party sponsorship. No "brought to you by" block or other paid placement appears in this issue.
Mapping against Ray Data Co
The sharpest connection is Thorsten Ball's line, relayed secondhand in this issue: a model that's wrong 10% of the time doesn't tell you which 10%, so you pay for the gap in retries. That's the exact premise behind RDCO's entire fresh-eyes-critic architecture (verify-vault-write, verify-strategic-output, verify-dispatch, station-critic) and the feedback_workflow_agent_output_integrity memory's seven factory failure modes — the model can't self-report its own error, so an independent gate has to catch it instead of trusting the producer's confidence. Second, the piece's "personal benchmark" habit (test each new release on one hard task you know well before switching your default) is already how the founder operates in practice: feedback_delegation_model_effort_pairing records the explicit 10-06 call to run everything on Opus 5.5 over Fable 5.1, and reference_claude_plan_max20x tracks the Max 20x plan itself as the vehicle for that kind of frequent re-evaluation. The "shoggoth strategy" debate (specialist front, generalist range behind it) is weaker — it's genuine career-positioning color relevant to project_credibility_for_phdata_sales, but the note shouldn't oversell it as a load-bearing RDCO insight; it's a useful frame, not a new decision.
Related
- [[2026-10-26-every-anthropic-worth-2-trillion]]
- [[2026-10-27-stratechery-anthropic-s1-cost-of-compute]]
- [[feedback_workflow_agent_output_integrity]]
- [[feedback_delegation_model_effort_pairing]]