News

What just shipped, what to read.

Model releases, paper summaries, ecosystem moves. Curated, opinionated, dated. Updated when something actually matters.

• Apr 28, 2026modelAnthropic

  Claude Opus 4.7 ships with 1M-token context window

  Anthropic's flagship reasoning model now ingests 1M tokens — comfortably the entire codebase of most products in one shot. Pricing unchanged.

• Apr 25, 2026ecosystem

  MCP one year on — what stuck, what didn't

  Model Context Protocol crossed 1k public servers in early 2026. The "USB-C for agents" pitch is mostly working. Where the rough edges remain.

• Apr 22, 2026business

  EU AI Act enforcement — what it means for your AI features

  Q1 2026 brought the first round of EU AI Act enforcement on high-risk systems. Five things product teams should check.

• Apr 15, 2026model

  Llama 4 ships open weights — and the math finally favors self-hosting

  Meta's Llama 4 70B and 405B are out. With API prices flat-to-rising, the breakeven point for self-hosting has shifted meaningfully.

• Apr 10, 2026paper

  Agentic benchmarks are still mostly broken — here's what isn't

  GAIA, AgentBench, SWE-bench are widely cited and widely gamed. A practical take on what to use for your own agent eval.

• Apr 5, 2026business

  Gemini 2.5 family — Google cuts pricing 30% across the board

  Gemini 2.5 Flash and Pro both saw 30% price cuts in April. Token economics are now genuinely competitive with Anthropic and OpenAI.

• Apr 2, 2026ecosystem

  RAG pop quiz: 73% of teams still use cosine similarity, 12% use hybrid

  A community survey of 250 teams running production RAG. The split between "default cosine" and "hybrid + reranker" is still huge.

• Mar 28, 2026ecosystem

  Voice-mode in 2025: lessons from a year of production deployments

  Speech-to-speech models are everywhere now. The teams that succeeded share three patterns. The ones that didn't share three failures.

• Mar 22, 2026model

  DeepSeek V4 distill — open weights with frontier reasoning

  DeepSeek's V4 distillation series brings frontier-model reasoning quality to ~30B-parameter open-weight models. Self-host changes again.

• Mar 15, 2026model

  Apple Intelligence: 3B on-device model now matches GPT-3.5

  Apple's on-device foundation model crossed an interesting line — small enough to fit, capable enough to be useful for many tasks.

• Mar 8, 2026paper

  100 failed agents — what we keep learning

  A retro on a year of production agent failures across consulting engagements. Top causes ranked.

• Mar 1, 2026ecosystem

  llms.txt — the emerging standard for AI crawlers

  Like robots.txt for AI training. Sites adopting it now will shape how future LLMs understand the web.

Subscribe to the newsletter (footer) to be told when something major drops.