Updates for 5/14

Ramp AI Index: Anthropic surpasses OpenAI in U.S. B2B adoption for the first time (34.4% vs 32.3%), having quadrupled its reach in 12 months

A year ago, the enterprise default was almost universally ChatGPT. Claude was well-regarded by engineers calling the API, but rarely appeared on procurement contracts or IT approval forms. The Ramp index is credible because it measures actual recurring payments by business customers — not signups or free-tier usage.

Vendor comparison decks can no longer skip Claude. Six months ago, 'just go with OpenAI' was a safe default; today that argument needs more support. That said, this shift reflects Claude's customer base broadening rather than OpenAI declining — many enterprises now run both. For individual users, ChatGPT still dominates awareness, so the day-to-day impact there is negligible.

"Claude for Small Business" launched with 15 agent-based workflow integrations for QuickBooks, PayPal, HubSpot and more, plus free training courses and a 10-city US workshop tour

Connecting Claude to QuickBooks or HubSpot required an engineer comfortable with connector setup and OAuth flows — not something a five-person shop has on hand. Claude Cowork launched last month for paid plan users, but setup complexity remained a barrier for non-technical small business owners.

This is Anthropic pushing past the enterprise tier for the first time. Invoice sorting, customer follow-up reminders, and payment reconciliation are now within reach for businesses without a dedicated IT team. The caveat: the rollout is US-focused, so international availability on the same integrations is not confirmed yet. If you already run QuickBooks or HubSpot, it is worth checking which workflows you are still doing manually.

Legal-focused Claude connector app launched covering transactional, employment, and litigation work — embeds into existing law firm workflows via connectors

The typical way lawyers used Claude was paste-and-prompt: drop a contract in and ask for a summary. Integrating Claude into repeatable legal workflows — contract review checklists, discovery sorting, employment agreement comparison — required rebuilding prompts each time, which made it impractical for routine operations.

Law firms and in-house legal teams that adopt this can start moving first-draft work in-house rather than outsourcing to outside counsel or paralegals. That said, legal liability for final decisions still sits with humans, so the value is faster drafting plus human review, not fully automated judgment. If you are evaluating AI vendor fit for a legal team, this narrows the gap between Claude and legal-specific tools like Harvey.

"Gemini Intelligence" announced for Android — cross-app automation (e.g. shopping list to cart), Chrome auto browse, Rambler filler-free voice transcription, and custom widgets; rolling out to Galaxy/Pixel from summer 2026

Android AI assistants have largely been within-app helpers — open a browser, maybe navigate to a site, but the actual transaction was still manual. Apple Intelligence had pushed into cross-app actions on iPhone, and Android was visibly behind on this front. The gap was especially obvious for task flows that span notes, browser, and commerce apps.

From summer, Galaxy and Pixel users will start experiencing what 'delegate this to my phone' actually means — not just 'set a reminder' but 'buy this.' For marketers and PMs thinking about mobile commerce, this changes how purchase intent converts: the AI layer in the device starts competing for the transaction. It is an announcement, not a launch, so nothing changes today — but the roadmap is now concrete enough to plan around.

Experimental Gemini-powered AI mouse pointer unveiled — captures visual and semantic context around the cursor, enabling natural-language shorthand without switching to a separate AI window

Working with AI has meant constant tab-switching: your actual work in one window, the AI chat in another. Copilot and Apple Intelligence both went with a side-panel approach — still a context switch. Full-screen computer-use agents could read the whole display, but they were heavy, latency-prone, and took a screenshot on each action.

Limiting the AI view to cursor proximity is a smart tradeoff: lower latency, less data exposure, and more surgical than full-screen capture. If this graduates from experiment to product, the finish-thought-then-switch-windows loop could largely disappear for power users. Most meaningful for designers working in Figma-type interfaces and engineers in IDEs. It is experimental today, so plan around it rather than act on it.

Launched developer platform to turn Notion into an AI agent hub — orchestration layer enables external data and agent integrations; users have already built 1M+ custom agents

Notion AI has been an all-in-one story — connectors to Slack, Drive, GitHub, but no way to plug in a third-party AI agent or orchestrate multiple agents from a Notion context. Wiring that up yourself meant building custom API integrations, which put it out of reach for non-technical teams.

For teams who already live in Notion, the docs and data hub can now become the coordination layer for AI workflows, not just a repository. PMs doing workflow design get a cleaner surface to experiment with multi-agent setups without needing a separate orchestration tool. The caveat: this is a developer platform launch, so no-code access follows later. Right now the value is clearest for engineers building agent-powered products on top of Notion data.

Incognito Chat added to Meta AI (rolling out to WhatsApp and Meta AI app) — conversation data is not stored persistently on Meta's servers

Meta AI conversations have sat on infrastructure shared with Instagram, Threads, and Facebook data. For sensitive queries — health, legal, personal matters — that made many users uncomfortable, and EU regulators had been watching closely. ChatGPT and Claude already had temporary-chat and memory-off options; Meta was visibly behind.

Topics that felt off-limits in Meta AI — medical questions, legal dilemmas, personal situations — now have a safer-feeling mode on WhatsApp, which is the primary messaging app for a huge share of global users. For enterprise use, the psychological barrier to discussing confidential matters via Meta AI lowers. Caveat: 'not stored persistently' needs careful reading — it likely means session-only, but the exact data handling policy should be verified before treating it as fully private.

Operating 46+ gas turbines at Mississippi data center without proper air permits; NAACP Clean Air Act lawsuit pending, and 19 more units added after suit was filed

AI data center energy use has drawn criticism for years, but actual illegal-operation lawsuits against AI labs have been rare. xAI expanded its Mississippi facility at speed, apparently skipping the permit process along the way. The NAACP filed suit under the Clean Air Act, and rather than pausing, xAI added 19 more turbines after the suit was filed.

For enterprises currently using or evaluating Grok API, this is now a vendor risk line item, not just a PR issue. ESG-focused procurement and legal teams need to assess this as an environmental liability in their AI supply chain. In a worst-case legal outcome there is non-zero risk of operational disruption to Grok API. For individual users the practical impact today is minimal, but enterprise buyers should track this case and have a contingency vendor in mind.

Uni-1.1 image model API launched — $0.04/image at 2048px, ranks 3rd overall on Arena leaderboard, includes built-in web search and reasoning, supports up to 9 reference images

Luma has been a video (Dream Machine) and 3D (Genie) company. They had no image generation offering, so product teams building with Luma for video or 3D still had to wire up a separate image API from Stability, Midjourney, or Imagen for static assets. Third place on Arena's leaderboard at 2048px is a strong debut for a first release.

Teams already using Luma for video or 3D can now consolidate image generation under the same API, reducing vendor management overhead. At $0.04 per image, it is competitive for high-volume use cases like e-commerce thumbnails and social content. Nine reference images for style consistency is a real advantage for brand content workflows. For anyone whose core work is photorealistic images or heavy video, the practical shift is smaller — but worth benchmarking against your current pipeline.