techThu 24 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

Claude discovers a potential new gene-editing system

Anthropic's molecular biology lab let Claude search 200,000+ enzymes in parallel—950 agents ran for 21 hours and spotted a repeating DNA pattern that looks like CRISPR (the gene-editing tool now used in medicine). Nobody knows yet what it does, but the structure suggests it could cut and paste DNA.

Jev in 25 lines of Python

HN's top story shows how elegant Jev's core is. Jev is the efficient small model everyone's been talking about—this breaks down the essence.

AI's shattering dev egos and that's the best part

Ryan Florence: you were never as smart as you thought, and that's liberating. The craft is shifting from writing code to orchestrating AI. People who can do that beat both old-school coders and raw AI.

Does code quality still matter?

Hot take: "AI is shattering dev egos, the best part of this industry." Counter: skilled people beat AI alone, and skilled people with AI beat both. Agreement: the game is who can orchestrate AI well, not who writes fastest.

Safari 27 ships MCP server for live debugging

Agents can now debug your web work in real time. Also landed: scroll anchoring (stops layout thrashing), customizable select elements, sizes="auto" for images.

Big announcements

Thursday, September 24, 2026 · Pergent2

Opus 5.5: 40% cheaper, way faster

Anthropic's new flagship Claude is 30% faster than Opus 5, costs 40% less ($4 input, $20 output per M tokens vs $5/$25), and beats Fable 5.1 on agentic work. Cache reads dropped 60%, which matters most for agents. Risky tasks (cyber, bio) reroute to older models.

OpenAI cuts frontier model prices ~50%

GPT-6 Sol ($2 in, $10 out) and Luna ($0.10 in, $0.50 out) undercut previous versions. Sol outperforms old OpenAI models and Opus 5 on automation, coding, and agent benchmarks. Both available now in ChatGPT Work for Plus/Pro/Business users.

ChatGPT Voice now uses plugins—email, calendar, Slack

Voice mode powers your tasks with GPT-6 Astra, Sol, or Luna models. Say what you need, it handles email, calendar, Slack, docs, sheets—all via voice on web and mobile.

Claude Cloud sessions: coding while away

Keep Claude Code running even when your laptop closes. Out of research preview, one-time credits: $100 for Pro, $250 for Max.

Qualcomm phones now run 30B-parameter models

Snapdragon 8 Elite Gen 6 and Extreme Gen 6 jump from 4B to 30B parameters—a 7x leap. Launching next year in Motorola, Xiaomi, OnePlus, Oppo, Vivo, and others.

What people are fighting about

Thursday, September 24, 2026 · Pergent3

Should plan mode die?

Claude Code team wants to kill plan mode—shift+tab to adjust effort levels instead. Models are good enough now. Plan mode diehards push back. The core argument: do you need to sketch before coding if the model is capable enough?

Claude Code reads agents.md with telemetry on

Privacy concern: Claude Code apparently reads your AGENTS.md definitions—but only when telemetry is enabled. It's by design but wasn't documented. Raises questions about what data flows during telemetry.

Web dev corner

Thursday, September 24, 2026 · Pergent4

Claude.ai got 3x faster in two weeks

Anthropic tuned performance with prompt engineering, UI updates, and caching. Blog post breaks down the techniques—if you're optimizing your own AI product, the playbook is here.

CSS nesting and :has() kill your Sass dependency

Native CSS nesting, :has() selector, and other modern features replace tools you've been using. You can retire Sass and similar toolchain dependencies.

From the newsletters

Thursday, September 24, 2026 · Pergent5

OpenAI's cheaper models change the math

GPT-6 Sol beats previous OpenAI models and Opus 5 on benchmarks while costing half as much. Shows companies read the trend: enterprises don't need frontier models for every task.

Research & open source

Thursday, September 24, 2026 · Pergent6

Worth a read

Thursday, September 24, 2026 · Pergent7

Opus 5.5 generates videos in 1 minute for $2

Instructional videos used to take weeks and agencies. Now one prompt, Opus generates the script and video. Changes everything about how we share technical ideas.

Writing code by hand is no longer economically viable

DHH's provocative take: pencils down. But the future of making software has never been brighter. Don't black pill this.