JadidJust-in AI Developments, Incidents & Deadlines
May 2026
SMTWTFS12345678910111213141516171819202122232425262728293031

AI news, Saturday, May 16, 2026

9 stories

TLDR

  • Notion becomes an AI agent hub (May 13) — Notion 3.5 ships Workers (hosted code runtime), an External Agents API (Claude, Codex, Decagon), and Database Sync; 1M+ agents already built by customers.
  • Colorado AI Act is effectively dead (May 7–9) — SB 189 passed both chambers and heads to the governor; the original June 30 compliance deadline is off the table thanks to both a federal enforcement stay and the lighter replacement bill.
  • Google I/O 2026 is this Tuesday–Wednesday (May 19–20) — Gemini 4 upgrades, Android 17, Gemini Omni video, and Project Astra updates all expected; watch for developer API and pricing changes.
  • SubQ 1M-Preview debuts with extraordinary (disputed) claims (May 5) — Miami startup Subquadratic claims the first fully subquadratic LLMLarge language modelA neural network trained on huge amounts of text that generates text, answers questions and summarizes.Anthropic ↗ with a 12M-token context windowHow much text a model can see and use at once while writing a reply.Anthropic ↗ at 1/5th frontier cost; AI researchers are demanding independent proof.
  • Meta's 8,000-person layoff goes into effect this Thursday (May 20) — first announced April 23 alongside $115–135B in 2026 AI capex; a reminder the wave is cresting this week.

Action Items

Things your engineering team should look at or do this week.

  • Block time for Google I/O May 19–20 — Gemini 4 and new tooling updates are expected; prior I/O events have preceded rapid Gemini API pricing and model-string changes. Watch the developer sessions for agenticDescribes AI systems where a model uses tools and decides its own steps instead of following a fixed script.Anthropic ↗ and multimodal API changes that affect your stack. (Google I/O 2026)
  • Reassess Colorado AI Act compliance work — If your team built out risk-management programs, impact assessments, or discrimination-auditing workflows for the June 30 deadline, you can pause those efforts: enforcement is stayed and SB 189 replaces the original law with a lighter notice-and-transparency regime. Track the governor's signature and the AG's rulemaking timeline for a new compliance horizon. (Law and the Workplace)
  • Evaluate Notion's External Agents API — If your team has Claude-based or Codex-based workflows on the roadmap, Notion now offers native out-of-the-box connectors plus a Workers runtime that requires zero infrastructure. Worth a 30-minute spike before building your own integration layer. (TechCrunch | Notion Releases)

Models & Tools

  • SubQ 1M-Preview (May 5) [Open Source] — Subquadratic, a Miami stealth startup, launched the claimed first commercial LLM built on a fully subquadratic sparse attention (SSA) architecture: 12M-token native context window, roughly linear compute scaling, and API pricing at ~1/5th of frontier models. The company raised $29M seed and is shipping an API, a CLI coding agent, and a search tool in private beta. Caveat: VentureBeat reports that several prominent researchers are publicly demanding independent third-party benchmarks before accepting the efficiency claims — treat as "promising architecture, verify before building on." (VentureBeat | Subquadratic Blog)

  • ZAYA1-8B by Zyphra (May 6) [Open Source] — Open-weightA model whose trained weights are published for anyone to download and run.NTIA ↗ MoEMixture of expertsA model made of many sub-networks, called experts, where a learned router sends each input to only a few of them.Shazeer et al. 2017 ↗ reasoning modelA model that works through hidden reasoning steps before its final answer. Those steps are billed as output.OpenAI ↗: 8B total parameters, ~760M active per tokenA small unit of data, such as part of a word, that AI models read and generate. Many AI services price by tokens in and out.NVIDIA ↗, Apache 2.0, and — notably — trained entirely on AMD Instinct MI300X GPUsGraphics processing unitA highly parallel processor that does the large matrix math behind training and running models.NVIDIA ↗ (1,024-GPU cluster), not NVIDIA hardware. Scores competitively with much larger open models on AIME, LiveCodeBench, and GPQA-Diamond. Weights on Hugging Face; free serverless endpoint on Zyphra Cloud. Relevant both as a capable small reasoning model and as proof that AMD-native training pipelines are now production-viable for frontier-class work. (VentureBeat | Zyphra | HPCwire)

  • Notion Developer Platform 3.5 (May 13) [Agents] [Customer AI] — Notion launched Workers (serverless hosted runtime for custom code, free in beta until August 11), an External Agents API with first-party Claude, Codex, and Decagon integrations plus a bring-your-own-agent path, and Database Sync for live pull from Salesforce, Zendesk, Postgres, and others. The company says 1M+ agents have been built by customers since preview. Notion's pitch: an orchestration layer where people and agents collaborate across tools and data, not just a note-taking app with AI sprinkled on top. (TechCrunch | Notion Releases)

APIs & Pricing

  • Google I/O 2026 (May 19–20): expect Gemini API changes (upcoming) — Confirmed topics include Gemini 4 model upgrades, agentic coding tools, and updates to Veo (text-to-video). Historically Google I/O has preceded near-term pricing and deprecation announcements. Audit your Gemini API usage and model strings before the week is out. (Google I/O 2026 | Tom's Guide preview)

Industry & Business

  • PayPal to cut ~4,760 jobs (~20%) over 2–3 years (May 5) [Hiring] — CEO Alex Chriss framed the move as "becoming a technology company again," with AI and automation expected to generate $1.5B in cost savings. Eliminating roles across support, operations, and non-AI engineering. Adds to the Q2 2026 AI-restructuring wave (Cloudflare, Cisco, LinkedIn, Coinbase) where growing revenue and headcount reductions are arriving in the same earnings cycle. (TechCrunch)

  • Meta's 8,000-person layoff goes into effect May 20 (announced Apr 23) [Hiring] — The 10% workforce cut — alongside cancellation of ~6,000 open roles — goes live this Thursday. Backdrop: $115–135B in 2026 AI capex, Llama infrastructure investment, and a stated goal of "smaller, faster, more agile teams." Severance: 16 weeks base + 2 weeks per year of tenure, plus 18 months COBRA. (CNBC | The Next Web)

Policy & Safety

  • Colorado AI Act replaced — June 30 compliance deadline off (May 7–9) [Policy] — A one-two punch: a federal magistrate issued an enforcement stay on April 27 (DOJ + xAI lawsuit challenging constitutionality), and the legislature passed SB 189 on May 7–9, which is now headed to Governor Polis. SB 189 strips out the original law's risk-management programs, annual impact assessments, and algorithmic discrimination duties, replacing them with a narrower notice-and-transparency framework. No new compliance deadline exists yet — it depends on the governor's signature and AG rulemaking. Enterprises should pause original-law prep work but maintain an AI-inventory as foundational compliance hygiene regardless. (Law and the Workplace | National Law Review)

  • ml-intern — An autonomous ML engineer that reads papers, hunts Hugging Face datasets, fine-tunes models in a sandbox, and ships production-ready code with minimal supervision. Rapidly climbing GitHub stars as teams look for "agent does the grunt work" patterns. (GitHub trending)

  • Hermes Agent (Nous Research) — Crossed 140K GitHub stars in under three months; currently the most-used agent on OpenRouter. Designed for reliability and self-improvement; built on NVIDIA RTX/DGX Spark. Worth benchmarking if you're evaluating open agent frameworks. (NVIDIA Blog)

Type to search · ↑↓ to move · enter to open