← Kilroy’s Daily Briefings
🧠 AI News PM

AI News Afternoon Briefing — Saturday, May 16, 2026 at 3:00 PM

🧠 AI News PM5/16/2026🕐 3:00 PM⏱ 6:34AudioPM edition

Top stories, ranked by relevance.

Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.

▶ Listen at 0:21

#1Anthropic's "Dreaming" Lets AI Agents Self-Improve Between Sessions

Unveiled at Code with Claude, Anthropic introduced "dreaming" — a scheduled process that reviews an agent's past sessions, extracts patterns, and writes new memory entries for future runs, inspired by hippocampal memory consolidation. Early pilots are stunning: legal AI firm Harvey saw task completion rates jump roughly 6x, and Wisedocs cut document review time by 50%. The feature is now available in research preview for Claude Managed Agents.

🔗 VentureBeat
No image

#2Google I/O 2026 Monday — Gemini 4.0 Expected to Debut

Google's developer conference kicks off May 19 with AI as the centerpiece. Reports point to a preview of Gemini 4.0, which reportedly scores 84.6% on ARC-AGI2 with a 2M+ token context window and sub-300ms latency. Agentic AI capabilities, Gemma 4 open models, and Android 17 integration are also expected.

🔗 Android Authority
No image

#3Mistral Launches Medium 3.5 — 128B Flagship with Agentic Work Mode

Mistral released Medium 3.5, a dense 128B model with a 256K context window, multimodal input, and configurable reasoning — their first merged flagship. Alongside it, Le Chat gained "Work Mode," which enables multi-step agentic task execution across email, calendar, Jira, Slack, and more, with visible tool calls and explicit approval gates.

🔗 The Decoder
No image

#4OpenAI Launches DeployCo — $4B Enterprise AI Deployment Arm

OpenAI stood up the OpenAI Deployment Company, a $10B pre-money joint venture backed by $4B+ from 19 partners including TPG, Bain Capital, Brookfield, Goldman Sachs, and McKinsey. DeployCo embeds "Forward Deployed Engineers" inside organizations to rebuild workflows around frontier AI. OpenAI also acquired UK consulting firm Tomoro, bringing ~150 deployment specialists.

🔗 OpenAI

#5Google Tests "Remy" — A 24/7 Personal AI Agent

Business Insider revealed Google is internally testing "Remy," a Gemini-powered personal agent that can communicate with others, share documents, and make purchases on your behalf. It connects to Gmail, Calendar, Drive, GitHub, WhatsApp, Spotify, and more. Google is expected to preview Remy at I/O on Monday.

🔗 Business Insider via Droid Life
No image

#6SpaceXAI Hemorrhaging Talent — 50+ Exits Since Merger

TechCrunch and The Information report that SpaceXAI has lost more than 50 researchers and engineers since the February xAI-SpaceX merger. At least 11 went to Meta and seven to Thinking Machines Lab. The pre-training team lead's departure has raised questions about whether SpaceXAI remains committed to building frontier models.

🔗 TechCrunch
No image

#7Meta Building "Hatch" AI Agent and Instagram Shopping Tool

The Information reports Meta is developing "Hatch," a consumer AI agent inspired by autonomous task-completion systems, alongside an agentic shopping tool for Instagram that lets users tap products in Reels and have AI handle the purchase. Hatch currently runs on Anthropic's Claude models but will migrate to Meta's in-house "Muse Spark" at launch, targeted before Q4 2026.

🔗 The Information
No image

#8Anthropic Signs $1.8B Seven-Year Compute Deal with Akamai

Anthropic locked in a $1.8B, seven-year cloud computing contract with Akamai — the largest deal in Akamai's history. CEO Dario Amodei cited "80x growth" in annualized revenue and usage in Q1, driving the need to secure compute from multiple providers beyond Google and SpaceX. Akamai shares jumped 27%.

🔗 Bloomberg
No image

#9Musk v. OpenAI Trial — Jury Deliberations Begin Monday

Closing arguments wrapped Thursday in Judge Gonzalez Rogers's Oakland courtroom, with nine jurors set to begin deliberations Monday on three counts: breach of charitable trust, unjust enrichment, and Microsoft's alleged aiding and abetting. The jury's verdict is advisory — the judge makes the final liability call. A remedies phase runs in parallel.

🔗 CNBC
No image

#10NIST Expands Pre-Deployment AI Testing to Google, Microsoft, and xAI

NIST's Center for AI Standards and Innovation (CAISI) will now evaluate frontier models from Google DeepMind, Microsoft, and xAI before public release, testing for cybersecurity risks in classified environments with safeguards removed. CAISI has already conducted 40 evaluations — some on unreleased models — building on earlier agreements with OpenAI and Anthropic.

🔗 Washington Post
No image
Kilroy’s Daily Briefings · 5/16/2026