Sarmadi AI Digest July 5, 2026 Updated 7:00 AM CT Today Archive Topics Saved Subscribe RSS

GPT-5.5 Codex hits reasoning-token clustering bug; Willison ships sqlite-utils 4 mostly from Claude Fable for $149

A widely-shared GitHub issue (293 HN) reports GPT-5.5 Codex is degrading due to reasoning-token clustering — a concrete performance-regression story that lands directly on the Zuckerberg 'agents haven't progressed as quickly' admission from Wednesday. Simon Willison shipped sqlite-utils 4.0rc2 mostly written by Claude Fable at a total token cost of $149.25 — a clean practitioner data point on cost-to-ship in the current model tier. TC confirmed the Alibaba Claude Code ban with more detail. Midjourney turned the table on Hollywood by demanding studios reveal their AI-usage details — an aggressive move given last week's Verge scrutiny of Midjourney's own medical pivot. A quiet US holiday follow-up day with heavy weight on community signal.

3 papers 7 news 3 sources ← Latest

News

8 items

GPT-5.5 Codex degrades; Willison ships sqlite-utils 4 for $149

A GitHub issue (293 HN) reports GPT-5.5 Codex is degrading due to reasoning-token clustering — concrete evidence of a mid-cycle regression on a heavily-used tier. Simon Willison shipped sqlite-utils 4.0rc2 mostly written by Claude Fable at a total token spend of $149.25 — a clean cost-to-ship number in the current tier.

News Hacker News

GPT-5.5 Codex reasoning-token clustering may be leading to degraded performance

OpenAI Codex issue (293 HN points): users report GPT-5.5 Codex degrading due to reasoning-token clustering.

Why it matters
  • First widely-shared reproducible performance-regression report on GPT-5.5 Codex.
  • Lands directly on Zuckerberg's 'agents haven't progressed as quickly' admission from Wednesday.
  • Materially affects the enterprise-agent story Amazon, Microsoft, and Anthropic all invested against last week.
  • Sets a new procurement signal: token-cluster degradation as a mid-cycle risk category.
productscodeevaluation
Source →
News Hacker News

sqlite-utils 4.0rc2, mostly written by Claude Fable (for about $149.25)

Simon Willison (55 HN points): sqlite-utils 4.0rc2 mostly written by Claude Fable at a total spend of $149.25.

total token spend $149.25
Why it matters
  • Clean end-to-end cost-to-ship number for a real OSS release built with a frontier model.
  • Data point for the 'high-intensity AI adopters +10.2% headcount' story from June 30 — cost is now known.
  • Establishes a public benchmark other maintainers will reference for AI-assisted release budgets.
codecommunityproducts
Source →

TC confirms Alibaba Claude Code ban; Midjourney demands studios disclose AI use

TC published its own confirmation of Alibaba's workplace Claude Code ban, adding detail to yesterday's Reuters report. Midjourney turned around and demanded Hollywood studios disclose the details of their AI use — aggressive positioning given last week's Verge scrutiny of Midjourney's own medical pivot.

News TechCrunch AI

Midjourney wants Hollywood studios to reveal the details of their AI usage

TC: Midjourney demands studios publicly disclose the details of their AI usage — aggressive counter-move mid-scrutiny.

Why it matters
  • Midjourney is playing offense while under Verge scrutiny of its own medical-scanner pivot — unusual timing.
  • Reframes the Hollywood-AI transparency story from studio-vs-model to model-vs-studio.
  • Sets the disclosure-standard fight the SAG-AFTRA renegotiation will now reference.
marketpolicyvideo-generation
Source →

Log-Is-the-Agent paper; Claude Design System; Google Independence commercial; Mistral explainer

A widely-shared paper argues 'The Log Is the Agent' — treating the transcript as the operational unit. A Claude Design System prompt hit HN. Google shipped a Fourth of July commercial imagining a Declaration of Independence drafted with AI help. TC published a Mistral explainer for readers wanting to know the OpenAI competitor.

Also today