AI Product Updates Daily — May 20, 2026

Google I/O dominated today's news cycle. The keynote ran nearly two hours and dropped more announcements than the past several I/Os combined — two new model families, a 24/7 agentic assistant, a redesigned app, a revamped search box, and a cascade of Workspace tools. Meanwhile Anthropic moved quietly but purposefully: Claude Managed Agents gained enterprise-grade privacy controls, Claude Code crossed 125K GitHub stars, and a new small-business package went live. Cursor shipped a coding model that trades pedigree for price. Here is everything that landed today.

Google: a two-hour keynote

Gemini 3.5 Flash — the new default

Google unveiled the Gemini 3.5 family at I/O and immediately deployed the first member. 1

Gemini 3.5 Flash is now the default model in the Gemini app and Google Search's AI Mode. Google says it is four times faster than peer frontier models on output tokens per second, while outperforming Gemini 3.1 Pro on key benchmarks. It is optimized for long-running agentic workflows. 2

Gemini 3.5 Pro, positioned as a top-tier reasoning and coding model intended to compete with Claude Opus 4.7 and GPT-5.5, is in testing and will roll out next month.

Gemini Omni — a world model for everything

Google also announced Gemini Omni, a multimodal world model built to accept any input (text, audio, images, video) and generate consistent, physics-grounded output. The first release, Gemini Omni Flash, combines Gemini's reasoning with Nano Banana image generation and Veo video generation in a single model. It is available today to paid Google AI Plus, Pro, and Ultra subscribers inside the Gemini app and Google Flow. Free access via YouTube Shorts and the YouTube Create app is expected later this week. 1

Google DeepMind CEO Demis Hassabis described Omni as "a pivotal step toward AGI" at the end of the keynote.

Gemini Spark — always-on personal agent

Gemini Spark is Google's most ambitious product launch today. It runs 24/7 on Google Cloud virtual machines and can execute tasks in the background even when a device is off. 3

Spark integrates with Gmail, Google Workspace, and more than 30 third-party services — including Adobe, Dropbox, and Uber — via MCP. Chrome integration comes this summer; additional MCP support follows in weeks. High-risk operations require explicit user approval. It rolls out next week to Google AI Ultra subscribers in the U.S. 2

Gemini app: Neural Expressive redesign + Daily Brief

The Gemini app was rebuilt from the ground up with a design language Google calls Neural Expressive — fluid animations, haptic feedback, vibrant colors, and a new typography system. Gemini Live voice mode is now integrated into the core UI rather than a separate mode. The redesign is rolling out globally on desktop, Android, and iOS. 1

Daily Brief is a new feature that pulls together a user's inbox, calendar, and tasks each morning, prioritizes them, and suggests next steps. It is rolling out today to Google AI subscribers in the U.S. The Gemini app has more than 900 million monthly users across 230+ countries and 70+ languages. 3

Google is expanding the search box to support long conversational queries, and adding support for image, file, video, and Chrome tab inputs. A background AI can track topics 24/7 and build custom dashboards or trackers for ongoing queries. 4

Ask YouTube — a conversational AI search layer that can surface specific video clips in response to complex queries — is open today to U.S. YouTube Premium subscribers.

Subscription pricing changes

Google adjusted its AI subscription tiers. The Ultra plan's entry price drops from $250/month to $200/month, and a new $99/month Ultra tier was introduced. The $99 plan includes five times the usage limits of the $20/month AI Pro tier, 20 TB of cloud storage, YouTube Premium, and early access to Gemini Spark. All Gemini tiers drop per-session prompt caps; usage is now billed against a compute quota that refreshes every five hours on a weekly total. 1

Enterprise: Google Cloud and Workspace

For enterprise customers, Google announced the Gemini Enterprise Agent Platform with a new Managed Agents API that lets developers run custom agents inside Google-hosted secure environments through a single API call. Antigravity 2.0 adds a desktop app and CLI for developers to build, manage, and orchestrate agents. CodeMender, an AI security agent developed at Google DeepMind, can autonomously identify code vulnerabilities and apply patches — currently in limited customer testing. 5

Google Pics — a Canva-like image creation and editing tool integrated into Google Docs, Drive, and Keep — is coming this summer for paid subscribers. Google also shipped a new AI Content Detection API in Agent Platform to identify AI-generated content from Google and other major models. 4

SynthID, Google's invisible AI watermark, is now detectable in Chrome and Google Search via right-click. OpenAI and several other third parties have adopted SynthID, broadening its coverage. 4

Anthropic: quiet week, steady agent build-out

Claude Managed Agents: MCP tunnels and self-hosted sandboxes

Anthropic added two enterprise privacy features to Claude Managed Agents. 6

MCP tunnels (limited research preview) let agents connect to private-network MCP servers — internal databases, private APIs, ticketing systems — through a single outbound connection from a lightweight on-premises gateway. No inbound firewall rules; traffic is encrypted end to end.

Self-hosted sandboxes (public beta) move tool execution into a customer's own infrastructure while Anthropic's servers handle agent orchestration, context management, and error recovery. Supported sandbox providers at launch: Cloudflare, Daytona, Modal, and Vercel.

Both features were announced this week; the earlier May 6 Managed Agents release had added dreaming, outcomes, and multiagent orchestration, with Harvey reporting a 6× jump in task completion after adopting those capabilities.

Claude Code: 125K GitHub stars

Claude Code, Anthropic's terminal-native coding agent first released as GA in May 2025, has reached 125,000 GitHub stars and 20,500 forks across 53 contributors. 7 The project recently deprecated the npm installation path in favor of platform-native installers (Homebrew, WinGet, PowerShell, curl), signaling a push to be a system-level CLI rather than a Node dependency. A plugin model lets it accept third-party extensions, including a PR-embedded code review tool.

Cursor: Composer 2.5 powered by Kimi K2.5

Cursor shipped Composer 2.5, a new AI coding model built on top of Kimi K2.5. Google says it was trained on 25 times the synthetic coding task volume of its predecessor and matches Claude Opus 4.7 and GPT-5.5 on coding benchmarks — at a substantially lower cost than either. The model is optimized for long-running tasks, multi-file work, and complex instruction chains. Cursor is offering double usage credits to existing users for the next week. 2

Also shipped today

Amazon Alexa received an AI podcasts feature. 2
GPT-5.5 Instant replaced GPT-5.3 Instant as ChatGPT's default model approximately two weeks ago, per changelog tracking. 8
NotebookLM (Google) updated its Mind Maps feature to support full prompt steering — users can now direct Mind Map structure via natural language. 9
Anthropic and OpenAI together account for 89% of AI startup revenue, per a report circulating this week. 2

Next up: Gemini 3.5 Pro is due next month. Gemini Spark's broader rollout and third-party MCP connections are expected within weeks. Watch for Claude Opus 5 signals as Anthropic's Code with Claude conference winds down.