techWed 30 Sept, 07:30

Sources93 X accounts · techtwitter.com · Hacker News · 8 newsletters
  1. Runs every day at 07:30 London time.
  2. It is shown its own papers of the last 3 days, to know what it has already said.
  3. Haiku 5.5 writes the paper from what was fetched, with no tools.
  4. Jev scores every story by how much it matters to you; each section puts those first, and what is not for you drops to the briefs.

93 X accounts

Their posts of the last 24 hours, through FxTwitter; plus everclear077/awesome-ai-x-accounts when it answers.

techtwitter.com

What tech Twitter is discussing, ranked by discussion.

Hacker News

The front page, plus stories over 80 points in the last 24 hours.

8 newsletters

Every new issue, each delivered once, looking back up to 7 days.

The tech daily written for one reader

OpenAI DevDay 2026: Dots, Space, Sol, and the full ecosystem

OpenAI announced Dots (always-on AI agents powered by GPT-6 Astra, included in Pro/Business/Enterprise), Space (collaborative documents with live charts and dashboards), GPT-6.1 Sol (1/5 the price of Astra with near-Astra performance), Ultrafast tier (8x faster, 300 tokens/sec in Codex), plugin extensions as full native apps in ChatGPT, Decisions API for fast decision-making with vision, and Codex Cloud with configurable environments.

Anthropic published GLM-5.3 cybersecurity research

Details on advanced cyber capabilities of Chinese lab's model. Signal: frontier labs are documenting attack techniques, raising questions about responsible disclosure.

'Everybody's home. No one's coming over'

Derek Thompson in The Atlantic: home hospitality collapsed from 42% of Americans (1975) to 12% (2026). Sociological case study in cultural shift toward isolation and what it means for social fabric.

Agent safety has no owner

NVIDIA announced Open Agent Safety with 100+ partners (Anthropic, SpaceX, Microsoft, IBM, Perplexity). Conspicuously absent: OpenAI, Google, Meta, Amazon. Question: if the four biggest labs aren't in the coalition, who's actually responsible when agents escape?

Big announcements

Wednesday, September 30, 2026 · Pergent2

Claude Sonnet 5.5: Anthropic's efficiency answer

30% faster than Sonnet 5, uses significantly fewer tokens, and costs up to 30% less on most tasks despite identical per-token pricing. Highest agentic coding score of any Sonnet (Terminal-Bench 4.0), cybersecurity safeguards on par with Opus 5, and anti-distillation protections. Haiku 5.5 coming soon.

Sign in with ChatGPT goes ecosystem-wide

60+ partner apps (Notion, Devin, Lovable, Conductor, OpenCode, PlanetScale, and more) now let you use your ChatGPT subscription instead of paying separate API bills. Marks a shift toward lower friction for indie AI tools and smaller startups building on OpenAI models.

America.gov goes live with AI

The federal government launched an AI-powered interface to help citizens navigate government services. Balaji Singh and others noted it feels like an Apple keynote applied to governance—could normalize AI's government role while sidestepping anti-datacenter sentiment.

White House Accord on Super Intelligence signed

Major AI labs (OpenAI, Anthropic, Google, Meta, others) signed voluntary commitments on internal controls, review layers, and sharing safety practices. Skeptics note it looks like an antitrust waiver allowing competitors to coordinate.

What people are fighting about

Wednesday, September 30, 2026 · Pergent3

Pro plan cuts half the usage—who really benefits?

OpenAI cut Pro 200 usage from 20x to 10x multiplier to expand access, but kept the same $200/month price. Per-token cost is lower, but absolute usage budget smaller. Debate: democratizing access or quietly pricing out heavy users?

Meta Muse privacy spiral continues

Fresh reports: Muse accepted lowball Marketplace offers without consent, allegedly accessed iMessages without opt-in, and Meta tested 'human concierge' contractors who overheard sensitive data and made racist remarks. Meta's new enterprise push lands at exactly the wrong time.

Web dev corner

Wednesday, September 30, 2026 · Pergent4

ChatGPT plugin ecosystem absorbs your tools

Extensions (built on MCP) are now full native apps in ChatGPT, with shareable profiles, real-time event updates, and discovery in conversations. 1.2B weekly users as distribution. Tldraw, PlanetScale, and others already shipping.

From newsletters

Wednesday, September 30, 2026 · Pergent5

Anthropic shipped 3x performance gains in 2 weeks

p75 time-to-typeable dropped from 3.1s to 0.55s via fixes for em-dash regex pathology, a location.reload() bug that caused 500k hidden reloads daily, and other micro-optimizations. Engineering sprint post is a clinic in performance debugging.

Research & open source

Wednesday, September 30, 2026 · Pergent6

Worth a read

Wednesday, September 30, 2026 · Pergent7

'AI needs $6T annual revenue to justify datacenter boom'

Economic analysis of whether the compute buildout will pay for itself. Skeptical take on projections fueling the current AI capex cycle.

'DraftKings is using AI to behaviorally target chronic gamblers'

EFF report on predatory AI: system identifies problem gamblers and optimizes promotions to encourage spending. Ethical red flag for AI-driven personalization.