Guillermo Rauch on the inference neocloud rush: Vercel AI Gateway token demand up ~40% MoM all year
Nov 18·discuss·save
levelsio: Turn on unattended security upgrades (and auto reboot) on your VPS
Matt Pocock: 70-minute watch-me-work deep dive building a 21,000-line PR with his agent skills
Cold email playbook from a founder who went $0 to $5M+ ARR in 18 months
Why niche SaaS no longer hides from big tech in the AI era
Nov 17·discuss·save
levelsio: How messing around on a vibe-coded flight sim game turned into $200K, a Cursor investment and Vibe Jam
A new pre-training trend: no LR decay phase
Google paper: recursive self-improvement at the harness level
How we doubled the users ChatGPT sends Revid in 10 weeks
Anthropic: one agent often beats a sprawl of specialized subagents
Prompt of the day: have the agent evaluate a new library for you
How the Claude Code team thinks about verification
The external sandbox pattern for code written by coding agents
Inside the Microsoft outage: a latent human-written bug
Nov 16·discuss·save
Open sourcing your SaaS is the new SaaS playbook
From tasks to goals: how agent design changes
How we do agent evals at Together AI (67% → 96% task success)
Muse's head of product on building the year's top AI consumer product
How to get Claude to design without AI slop: 5 takeaways from Nate Parrott
My top 10 eval papers of 2026
Gergely Orosz x Satya Nadella: software engineering when code is cheap and verification is hard
Matt Pocock: "Nuance tends to lead to slop" - design APIs and checks that remove nuance from agent decisions
A brief history of AI coding since Nov 2025
Guillermo Rauch: xeet, a pixel-perfect X post renderer built as an agent-maintained 'autorepo'
Turn off auto memory in your coding agent harness
The AI ad stack that produced $1.4M ARR for a Series A startup
Nov 15·discuss·save
The best agentic systems are full of boring deterministic steps
Young founders listen too much to investors and not enough to users
Understanding positional embeddings in modern LLMs (RoPE from scratch)
How to pick a model: Opus 5.5 medium as a daily driver
Nov 14·discuss·save