techThu 1 Oct, 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

Gemini 4 Argon: Google's new frontier model lands with 1M token limit

Google announced Gemini 4 Argon, a frontier-tier model built for complex, long-horizon workflows across coding, enterprise knowledge work, and cybersecurity defense. It's rolling out to trusted cyber defenders starting today at $2 per million input tokens and $10 per million output—a fifth the price of its Astra model—with broader availability coming soon.

Anthropic GLM-5.3: cyberattack autonomy raises alarm

Anthropic published research showing GLM-5.3 can autonomously build cyberattacks and exploits, highlighting how advanced open models are becoming a dual-use security risk.

"You said no MCP": How platform gatekeeping kills the agent ecosystem

When Figma (and soon maybe Amazon, Reddit, others) restrict agent access to their APIs and data, every company starts playing gatekeeper. The result: agents lose optionality and the open ecosystem fragments. HN's most-discussed story (630 points).

Token spend limits: what happens when you hit your cap?

A heated thread on X from thdxr blew up (1,404 likes, 417 replies) asking: when your company caps monthly AI token spend, do you flip back to writing code by hand, or just stop working? Responses revealed a real affordability crisis for solo devs and small teams watching their budgets disappear as models get pricier and more capable.

Netlify Edge Functions get 5x faster startup with Firecracker

Netlify rebuilt Edge Functions on Firecracker MicroVMs instead of V8 isolates, achieving 5x faster startup time (p50 648ms) and 6x faster for TTI. The move also unlocked snapshots in beta, letting serverless functions start from a saved state.

Big announcements

Thursday, October 1, 2026 · Pergent2

Cognition hires Factory AI's CRO; confidentiality accusations fly

Cognition brought on Chris Degnan (ex-Snowflake CRO) as its own CRO hours after Factory AI CEO Matan Karush accused him on X of sharing confidential business details with Cognition while still on Factory's board. Cognition flatly denies the allegation and says Degnan never shared any Factory info.

Claude.dev: Anthropic launches developer hub with engineering deep dives

Anthropic opened Claude.dev as a home for developers building with Claude, with engineering guides, API documentation, tips from the team, and a few easter eggs. The site showcases how Anthropic made the Claude web app 3x faster, cutting time to first keystroke from 3.1 seconds to 0.55 seconds.

Flow raises $50M Series B for hardware engineering agents

Flow, a platform that uses agents to automate repetitive CAD and engineering work, closed a $50M Series B at $750M valuation led by Valor and Atreides Capital, with participation from Sequoia and other investors. Customers including Anduril, Joby, and Stoke Space are using it to accelerate hardware builds.

What people are fighting about

Thursday, October 1, 2026 · Pergent3

Figma blocks MCP, sparking platform gatekeeping fears

Figma restricted Claude from accessing design plugins via MCP, forcing builders to apply to an allowlist. The move triggered ecosystem backlash: "I envisioned an open ecosystem" said MCP creator David Hsu. Hacker News essay "You said no MCP" became the day's most-discussed post (630 points), arguing that if every platform gates access to their data or APIs, the agent ecosystem collapses.

White House AI safety accord relies on honor system, no teeth

The White House Accord signed by Nvidia, Anthropic, OpenAI, Meta, and Google pledges "layers of controls" on model safety but has zero enforcement—just a promise to self-regulate. Critics argue that without law and regulation, companies with financial incentives will ignore safeguards as AI approaches public market debuts.

Web dev corner

Thursday, October 1, 2026 · Pergent4

GitHub ditches CSS-in-JS, cuts server-side render time 55%

GitHub spent three years replacing styled-components and CSS-in-JS with CSS Modules on github.com, shipping more CSS but cutting server-side render time by 55%. CSS Modules scope styles per file with no runtime cost, letting bundlers do the work at build time instead.

From the newsletters

Thursday, October 1, 2026 · Pergent5

OpenAI launches GPT-6.1 Sol at one-fifth Astra's price

OpenAI rolled out GPT-6.1 Sol, delivering "near Astra intelligence" at $2 per million input tokens, $10 per million output (vs Astra's $10 in, $30 out). It's built for agentic coding, computer use, and professional work.

ChatGPT Space: collaborative workspace for teams and AI

A new collaborative workspace (like a shared Google Doc) lets you, your teammates, and ChatGPT work together. Everything shared with and created in ChatGPT stays in one place.

Ultrafast tier: OpenAI offers 8x Codex speed for $500/month

OpenAI created a high-end $500/month tier offering Ultrafast, delivering up to 8x speed in Codex using GPT-6 Astra and 6x speed in the API. But the $200/month Pro plan cuts Codex usage from 20x to 10x the Plus allowance.

EmDash 1.0: Cloudflare's open-source WordPress alternative

EmDash, Cloudflare's CMS for Astro, reached 1.0 and now ships with an MCP server so agents can set up sites out of the box. Plugins can be submitted to the registry.

Anthropic made Claude's web app 3x faster in one sprint

Anthropic cut Claude's time-to-first-keystroke (75th percentile) from 3.1s to 0.55s by tracking micro-performance issues: layout shifts too small for CLS to flag, a :root:has() selector costing 24ms per DOM change, and Chrome pre-rendering at the wrong height.

Research & open source

Thursday, October 1, 2026 · Pergent6

Worth a read

Thursday, October 1, 2026 · Pergent7

"The AI Race Just Got Awkward": Model wars heating up

Gemini 4 Argon undercuts Astra on price; Claude takes back the performance crown from GPT-6; open models close the gap. It's a fierce competition now, and nobody's sure who wins.