“Welcome to April 17, 2026” — The Innermost Loop

Why this is in the vault

Daily Wissner-Gross digest with hard signals on the Claude Opus 4.7 release confirmation, an internal Anthropic poll that ~1/3 of staff expect Mythos to replace entry-level engineers/researchers within three months, federal procurement of Mythos, OpenAI’s Codex update positioning the IDE as a coworker (a direct echo of Anthropic’s Cowork), and hyperscaler capex now exceeding the inflation-adjusted Apollo+Interstate+Marshall combined. These are operational frontier markers for RDCO, not generic AI news.

Bias / sponsorship check

No paid sponsors. Two standard “Subscribe for free” CTAs. Tone is the usual dry-satirical pro-acceleration framing; closing aphorism: “ships point releases faster than civilization can debug itself.” No self-promo of Solve Everything. No disclosed positions.

Core argument

A single-day sweep loosely organized around “the Singularity now ships on a schedule” — i.e., frontier capability is now a release cadence with consequences arriving on calendar time.

Frontier model release cadence. Claude Opus 4.7 confirmed shipped, framed as midpoint between 4.6 and the unreleased Mythos Preview. Internal Anthropic poll: nearly a third of staff expect Mythos to replace entry-level engineers/researchers in three months. White House OMB is routing Mythos into major federal agencies “in the coming weeks,” explicitly choosing acknowledged cyber risk over non-adoption. OpenAI countered with GPT-Rosalind, a frontier reasoning model purpose-built for biology / drug discovery / protein engineering.

Automation generating exhaust + AI-into-defense pipelines. NIST is restructuring CVE handling after a 263% jump in vulnerability reports (2020-2025) driven by AI submissions, narrowing triage to known-exploited and federally-relevant. Google reportedly negotiating to deploy Gemini in classified Pentagon environments. Boston Dynamics’ Spot now runs on Gemini Robotics-ER 1.6.

OpenAI Codex update — the IDE-as-coworker pattern. OpenAI shipped a Codex update that operates the user’s computer alongside them and remembers preferences, framed by Wissner-Gross as the IDE moving from “autocomplete to coworker” — a direct counter to Anthropic’s Cowork. Defunct startups are being liquidated for their Slack archives, Jira, and email threads as premium training data (“reincarnating failed companies as weights”).

Silicon layer compounding. TSMC expects 30%+ revenue growth this year. Cerebras filing IPO at $35B+, anchored by a $20B three-year OpenAI compute deal that includes warrants scaling with spend (collapsing customer/owner distinction). xAI becoming a cloud provider; Cursor reportedly training Composer 2.5 on tens of thousands of xAI GPUs.

Sensorium-as-API. Focused 300-kHz ultrasound to the olfactory bulb induces artificial smells with no cartridges. Sabi’s thought-to-text EEG beanie reads internal speech to device. South Korean group found a remotely controllable in vivo gene switch (Cyb5b as EMF sensor). Project CETI: sperm whale codas pattern like human vowels phonologically.

Capital + geopolitics. Alphabet poised for ~$100B from SpaceX IPO via residual 5% post-xAI-merger. Hyperscaler capex has surpassed inflation-adjusted Apollo + Interstate + Marshall combined at equivalent project age. Taiwan market cap crossed $4T (overtook UK). UK asking households to consume more during renewable peaks (abundance choreography vs. conservation rhetoric). US established a 4,000-acre Luzon SEZ with diplomatic immunity and US common law for China-proof automated supply chains. Snap cutting 16% to chase AI margins; Myseum doubled on AI pivot.

Tensions beneath the boom. Following the Molotov attack on Sam Altman’s house, OpenAI policy chief Chris Lehane warned doomers are “playing with fire.” White House vowed to investigate 10 US scientists/engineers/military leaders recently missing or dead.

Mapping against Ray Data Co

Opus 4.7 confirmed shipped + positioned as a midpoint to Mythos. Confirms yesterday’s signal in 2026-04-16-innermost-loop-welcome-apr-16 was correct on timing. RDCO is already running on it (model banner). No immediate action required, but every skill that assumed 4.6 behavior should be loosely re-spot-checked, and the “Mythos as next horizon” framing matters for content-calendar planning — if Mythos ships in weeks, any newsletter angle that hinges on current capability ceilings has a short shelf life.
~1/3 of Anthropic staff expect Mythos to replace entry-level engineers/researchers in 3 months. This is a hard signal to bring into the abundance-flywheel / muddle-vs-rails writing. It is a labor-substitution prediction made by the people building the substitute, on a calendar measured in single-digit months. Strong Sanity Check candidate: “the inside view of when entry-level disappears.”
OpenAI Codex update = IDE-as-coworker, direct counter to Anthropic Cowork. This validates the architectural bet behind RDCO’s /check-board autonomous loop and skill harness — the “agent that operates your computer alongside you and remembers your preferences” pattern is now table stakes across both labs. RDCO’s harness differentiates on vault-grounding and channel-routing, not on “having an agent.” Worth flagging for the moat-when-the-platform-catches-up Sanity Check angle already noted in the Apr 16 file.
Defunct startups liquidated for Slack/Jira/email as training data. Direct relevance to RDCO’s content-as-a-product positioning: the operational artifacts of a company are now an asset class. Implication for any client work — hygiene around what gets written in Slack/Jira changes when those archives have a downstream training market. Also a fresh angle for Sanity Check: “your dead company is someone’s pretraining set.”
Cerebras IPO with OpenAI warrants scaling with spend. The customer-becomes-owner pattern — same shape as Microsoft/OpenAI but more explicit. Worth tracking as the financial-architecture template for the AI buildout; useful as a “follow the money” prompt in the compute-economics thread.
Hyperscaler capex > Apollo + Interstate + Marshall combined (inflation-adjusted, equivalent age). The single most quotable infrastructure-scale stat to land in the vault this week. File as a reusable opener line for any abundance-flywheel essay.
NIST CVE restructuring under AI-driven submission load. Operational signal that AI-generated noise is now restructuring the institutions that absorb it. This is the same pattern as content moderation under generative spam — generally relevant for any RDCO client that runs intake/triage workflows, and a useful concrete example of “AI breaks the bureaucracy that AI then has to guard.”
UK demand-shaping for renewable peaks (consume more, not less). Inverts the conservation frame into abundance choreography. Direct fit with abundance-flywheel. Quotable.
Luzon SEZ with US common law + diplomatic immunity. Geopolitical-supply-chain signal — worth filing alongside any prior notes on Taiwan/China-proofing for the substrate layer.

2026-04-16-innermost-loop-welcome-apr-16 — yesterday’s issue confirms the Opus 4.7 release window predicted there
2026-04-13-innermost-loop-welcome-apr-13
2026-04-12-innermost-loop-singularity-immune-response
2026-04-09-innermost-loop-mythos-arrival-compute-scissor
2026-04-08-innermost-loop-mythos-glasswing
abundance-flywheel
muddle-vs-rails
compute-economics

Copyright note