AI Toolchain Power Shift | AI Field Notes #30

AnalysisGoogle unveiled Gemini 3.5 Flash at I/O on May 19, claiming it outperforms 3.1 Pro on coding, multimodal, and autonomous-task benchmarks while running 4x faster than competing frontier models at roughly one-third their cost. The model rolls out immediately in the Gemini app, Google Search, and the Gemini API. Alongside it, Google announced Gemini Spark, a personal AI agent that runs 24 hours a day on Google Cloud virtual machines even when your device is off, capable of working across Google Workspace, third-party apps, and the web. Spark launches first for Google AI Ultra subscribers next week.

AnalysisAnthropic bought Stainless on May 18 for over $300 million, removing a core piece of developer infrastructure from its biggest rivals. Founded in 2022 by former Stripe engineer Alex Rattray, Stainless automates SDK generation across Python, TypeScript, Kotlin, Go, and Java from API specifications. OpenAI, Google, Cloudflare, Replicate, and Runway all relied on it. Anthropic plans to wind down all hosted Stainless products, though existing customers retain rights to SDKs already generated. The strategic read: Anthropic has turned developer toolchain control into a competitive moat, forcing rivals to rebuild from scratch or migrate on their own.

AnalysisAnthropic expanded Claude Code's capacity on May 6, doubling 5-hour usage limits for Pro, Max, Team, and Enterprise plans and removing peak-hour throttling that had been slowing the tool during busy periods. The company attributed the change to a compute deal with SpaceX's Colossus 1 data cluster. Removing throttling is the more meaningful change: it eliminates the unpredictable mid-session pauses that forced developers to restart long autonomous runs, a problem that had made Claude Code inconsistent for multi-hour tasks where a pause can derail an entire run.

AnalysisCursor shipped Composer 2.5 on May 18, matching Anthropic's Opus 4.7 and OpenAI's GPT-5.5 on coding benchmarks while pricing at $0.50 per million input tokens and $2.50 per million output tokens, roughly one-fifth of what the frontier models cost direct. The release adds cloud agent dev environments, eliminating the need to run local containers, plus Microsoft Teams integration and a Build in Parallel feature for simultaneous agent threads. Composer now functions as an orchestration layer across multiple underlying models rather than a wrapper around a single one.

AnalysisGoogle's Antigravity 2.0, released at I/O, expands the agent-first development environment into a full platform: desktop app, CLI, SDK, and enterprise deployment through the Gemini Enterprise Agent Platform. New capabilities include dynamic subagents for parallel workflows, scheduled background tasks, native voice commands, and one-click deploy to Cloud Run with Firebase integrations. Separately, Google proposed WebMCP, an open standard letting developers expose tools directly to AI agents through browsers, with an experimental origin trial starting in Chrome 149. Both moves treat agent connectivity as platform infrastructure rather than product features.

AnalysisOpenAI launched a self-serve Ads Manager for ChatGPT on May 5, removing the $50,000 minimum spend that had confined advertising to large agencies. Small businesses and mid-market companies can now buy cost-per-click campaigns directly, with agency partners including Dentsu, Omnicom, and Publicis, and ad-tech integrations with Adobe, Criteo, and StackAdapt. Ads appear only in free and Go tiers; paid subscribers (Pro, Business, Enterprise) see none. OpenAI targets $2.5 billion in ad revenue this year and $100 billion annually by 2030. The company says advertiser data is never mixed into user conversations.

AnalysisPope Leo XIV will publish 'Magnifica Humanitas,' his first major teaching document, on May 25. The encyclical addresses artificial intelligence and human dignity. The Vatican announced that Christopher Olah, co-founder of Anthropic, will speak alongside the Pope at the Synod Hall during the presentation. Leo has compared AI's societal impact to the industrial revolution and is expected to situate the technology within Catholic social teaching on labor, justice, and human development. He has previously warned priests against using chatbots to write homilies and expressed concern about AI's effect on children's cognitive development.