techMon 2 Nov, 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

Bun performance

Bun says redesign, don't cache

The post drew 2,595 likes and 105 replies.

Jarred Sumner says he tells Claude to rewrite slow code with flat arrays and data-oriented design, and to delete caches instead of adding them. He argues that checking whether a cache is still valid can cost more than the work it saves, and stale caches can be a security problem.

Type ultrathink when a task gets hard

Addy Osmani says typing ultrathink in Claude Code raises effort for the rest of that turn, useful when a task turns out harder than expected.

Server security

Lock down server ports

A quick stopgap, he says, is turning off MPTCP on your server.

Pieter Levels says any Linux box can be taken over through one open TCP port, so all your VPS ports should be closed. He reaches SSH through Tailscale, a private network, and web traffic through a Cloudflare Tunnel.

Local models

Local AI vs. the cloud

Rauch's post drew 2,356 likes and 251 replies.

Guillermo Rauch, Vercel's CEO, says you should own your intelligence, your harness, your data and your agents, and that local models need to keep spreading. Teknium, who builds the Hermes agent, reposted a worry about cloud personal assistants and data privacy, and says he keeps his AI brain swappable.

Claude Code

Claude Code edits a vlog

Levelsio fed it 14 clips and it made a finished vlog with only a few tweaks.

Pieter Levels says Claude Code, running Opus 5.5, cut 14 clips into a 10-minute vlog. It transcribed speech with a local Whisper model, made contact sheets so Claude could see each clip, and used a voice-matching model to tell his voice from others.

State management

Redux may be back

The post drew 941 likes and 212 replies.

Matt Pocock says a Redux-style framework could make a comeback, since keeping state in one place gives AI agents fewer ways to go wrong. In a follow-up he admits state always lives in several places, like the URL and the server, so he thinks you need something like Effect too.

Big announcements

Monday, November 2, 2026 · Pergent2

Retrieval models

Particle-1 cuts search costs

The numbers come from a LongMemEval test run that is posted publicly on GitHub.

Garry Tan of Y Combinator says Particle-1 matches much larger models at agentic retrieval, the step where an AI agent looks up the facts it needs. It does this at 1/23 of the cost.

What people are fighting about

Monday, November 2, 2026 · Pergent3

AI evaluation

AI benchmarks under fire

Husain says most people haven't caught on yet, but they will.

Hamel Husain says AI benchmark companies run a business much like Gartner's vendor rankings, credit rating agencies and US News college rankings, all of which are paid by the things they rate. LlamaIndex's Jerry Liu, meanwhile, claims its LlamaParse Max Effort mode beats frontier agent setups on accuracy and cost.

AI research

LeCun corrects the record

The reply is in French, and it corrects a post about his role at Meta.

Yann LeCun says he founded FAIR in 2013, left its lead role in 2018, and was never involved in large language models, either in research or in product work. It comes as a viral post claims his proof shows a 109-million-parameter model can beat one with 72 billion.

AI pricing

Cheap AI plans won't last

The post drew 2,384 likes and 181 replies.

Amjad Masad, who runs Replit, says anyone enjoying a subsidized AI subscription should get ready to get rug-pulled. He doesn't name a company, so it isn't clear who he means.

Social media

X's AI reply problem

Matt Pocock's post got 528 likes and 131 replies.

Matt Pocock says AI replies on X are astonishingly bad, so he now looks for real profile pictures to tell people from bots. Pieter Levels says X seems to have removed the Hide reply button, so he can no longer hide or block AI replies. thdxr, of OpenCode, says he is fighting back with his own anti-bot reply system.

Web dev corner

Monday, November 2, 2026 · Pergent4

Security incident

Trigger.dev secrets exposed

Environment variables are where apps keep API keys and database passwords, so this matters.

Trigger.dev, a service that runs background jobs, says customer environment variables were exposed. Its incident post is on the company blog and got 66 points on Hacker News.

Linux TCP bug

Vercel finishes kernel patch

The bug is a use-after-free in Linux's IPv4 TCP networking code.

Vercel says it has finished applying Linux kernel security updates to fix CVE-2026-98432. This follows up yesterday's story on the bug, which got a 9.8 severity score, near the top of the scale.

Supabase

Postgres gets a new engine

Supabase says OrioleDB scales almost linearly with CPU cores.

Deepthi Sigireddi explains OrioleDB, a storage engine that removes bloat and vacuum, and Multigres, which adds high availability (a standby that takes over if the main server fails), automatic failover and connection pooling. Sharding is coming later.

Postgres

Postgres 19 slips over AI bugs

Haas is a long-time Postgres developer, and his post has 300 points and 198 comments on Hacker News.

Postgres 19, the next big version of the open-source database, is running late. Robert Haas says the reason is that many more bugs are being found now, and he blames AI for the flood.

Research & open source

Monday, November 2, 2026 · Pergent5

arXiv

A HackerTube for papers

It aims to surface papers people are talking about, not just the newest ones.

omarsar0 says he took Karpathy's HackerTube prompt and similar data sources, then asked his agent to build an arXiv version. It now holds more than 1,000 papers.

Agent teams

Minecraft agents drown in red tape

The Hacker News thread on it has 78 comments.

Tom's Hardware reports that 50 agent teams, let loose in one Minecraft world to build a civilization, fell into paralyzing bureaucracy and failed. It is a cautionary tale for anyone building multi-agent setups.

Worth a read

Monday, November 2, 2026 · Pergent6