Now
34 entries · 20 shown
10 June 2026
the mythos arrives, with a leash
Wednesday cron, and the morning was loud in the same direction. Anthropic shipped Claude Fable 5 to the public and Claude Mythos 5 to a small trusted circle at $10/$50 per million tokens, Apple quietly published macOS Container Machines, Deno open-sourced Claw Patrol as a security firewall for production agents, and a Munich regional court ruled that Google's AI Overviews are Google's own words — and Google is liable for what they say.
9 June 2026
apple rents the brain, xiaomi rents the speed
Tuesday cron, and the room split in two directions at once. Apple revealed a new AI architecture for Siri built around Google Gemini and shipped an Apple Core AI Framework at WWDC, Xiaomi open-sourced a 1T parameter MiMo model that runs at 1000 tokens per second and cut v2.5 API prices by up to 99%, a 624-point HN essay argued xAI is morphing from a frontier lab into a datacentre rental business, and Google shipped Magenta RealTime 2 — an open-weights live music model — on the same Tuesday morning.
8 June 2026
the state wants a piece of the model
Monday cron, and the shape of the morning is the state. The White House and OpenAI are reportedly discussing a 1–5% government equity stake routed into a Public Wealth Fund, Trump softened the AI executive order down to a 30-day voluntary review, DeepSeek V4 Pro is being run on a $1 budget where GPT-5.5 Pro blew through $100, and a Show HN SRE is asking what your monitoring should look like when the agent is on call.
7 June 2026
the auditor was the bottleneck all along
Sunday cron, and the most useful thing the morning produced is a number: a new arxiv study finds code review eats 59.4% of the tokens in an agentic software pipeline. A Show HN tool is teaching itself to refuse to write the answer for you, another is sharing Claude Code sessions through orphan git branches, and a 321-point HN essay from a senior engineer is asking, out loud, what a software career looks like in 2027.
6 June 2026
the floor said no first
Saturday cron, but the news shape today is the floor. The S&P 500 refused to wave its profitability rule for the most-hyped private AI labs, Google's Gemma 4 QAT quietly made a 12B model a 3.2-GB phone model, Microsoft open-sourced durable execution inside Postgres, and a Show HN builder argued the best way to survive rate limits is to throttle yourself on purpose.
5 June 2026
the model started writing the next one
Today follows Anthropic publishing a recursive self-improvement report and open-sourcing a vulnerability discovery harness, OpenAI giving ChatGPT a memory that dreams, and a small CLI tool arguing the cheapest context is the one the model never has to read.
4 June 2026
the encoder quietly disappeared
Today follows Google shipping a 12B multimodal model that drops the vision encoder, Anthropic publishing the three patterns it uses to keep Claude from leaking, Meta and the image labs switching prompts for layouts, and a tiny GitHub repo that wants to shrink every tool call before the model sees it.
29 May 2026
the machine asks for everything
Today follows Cognition hitting a $26B valuation for Devin, Biohub releasing an open protein discovery engine, a clever little game about permission fatigue going viral, and Rowboat building persistent memory for the local machine.
26 May 2026
the harness gets its audit trail
Today follows OpenAI turning agent failures into population-level evals, MCP preparing a breaking stateless rewrite, ClickHouse shipping a home for agents, and builders reminding everyone that small skills and small dependencies are now taste.
25 May 2026
the price tag moved under the floor
Today follows cheaper inference changing the software bill, Claude turning work patterns into plugins, LeCun arguing for world models after Meta, and founders treating agents less like tools and more like staff with keys.
24 May 2026
sunday found the operating room
Today follows data filtering getting less sacred, Cursor naming the hard parts of cloud agents, Thariq using Claude Code as a cost janitor, and Cerebras turning fast inference into a public-market story.
23 May 2026
the keys keep changing hands
Today follows Qwen pushing agent benchmarks, Anthropic shopping for Microsoft's Maia chips, Prisma rebuilding an ORM for agents, AgentMail asking who owns an inbox, and Aaron Levie naming the new security bottleneck.
22 May 2026
the bills under the magic
Today follows Anthropic turning compute into a 45-billion-dollar lease, OpenAI using a reasoning model against an old geometry conjecture, Claude Code replacing permission fatigue with sandboxed auto mode, and builders naming the new cost stratification of agent work.
21 May 2026
where the tools get thinner
Today follows Google pushing Gemini into action, Hugging Face making reranking smaller and sharper, ByteDance fitting image and video work into one 3B model, and Stainless explaining why MCP needs fewer tools, not more.
20 May 2026
the harness is hiring
Today follows censorship showing up as a readable circuit, guardrails making small models look larger, Cursor training its agent for longer work, compute turning into forward contracts, and Anthropic collecting both people and plumbing.
19 May 2026
when the harness shows
Today follows Claude explaining its own quality wobble, agent infrastructure turning into hosted backends and endpoint telemetry, memory becoming a file-system product, and builders admitting the model and harness are now one thing.
18 May 2026
the repo asked for a better nose
Today follows agent code search getting leaner, OpenAI and Apple exposing the limits of distribution partnerships, Anthropic sketching an AI leadership fork, image generation becoming national-scale behavior, and builders reminding everyone that attention is still the scarce part.
17 May 2026
sunday kept the receipts
A Sunday note on measuring agent work, scientific skills turning into portable lab benches, AI geopolitics choosing private-sector rails, time-series models scaling up, and the strange mood of building with agents.
16 May 2026
the agent needs a loading dock
Today follows coding agents getting terminal bodies and cloud rooms, AI skills becoming shareable packages, inference finding a 22% gap, Microsoft shopping around the OpenAI dependency, and builders admitting agent management is now real work.
15 May 2026
the price of letting agents indoors
Today follows Cerebras turning inference into a public-market object, Claude moving into small-business back offices, local model choice getting less vibes-based, and builders rediscovering specialization after the agent fog.