The AI Enthusiast.

Anthropic looks like the premium lane.

This was the first week where Anthropic's story felt less like "better model" and more like "better business." Opus 4.7 sits at the top of the stack, pricing stayed flat, and the company paired product momentum with a blunt message: we are financing more compute because demand is already here.

Anthropic said on April 6 that run-rate revenue had moved past $30 billion and that customers spending more than $1 million annually doubled in under two months. On April 20 it added an AWS deal worth more than $100 billion over ten years, securing up to 5 gigawatts of capacity and nearly 1 gigawatt before the end of 2026. That is not experimentation behavior. That is "we know where the demand is coming from" behavior.

Opus 4.7 reinforces the same point. Anthropic kept pricing at $5 per million input tokens and $25 per million output tokens, then positioned the model around longer-running coding, vision, and finance-heavy work instead of general chatbot theater. The signal is clean: premium workflow spend beats breadth if the outputs matter enough.

Claude Code stops babysitting.

The real UX shift is not a benchmark. It is Claude Code becoming harder to interrupt. Auto Mode, routines, ultrareview, and better usage visibility move the product from "assistant in a terminal" toward "agentic runtime with opinionated controls."

The more important strategic read is that Anthropic is teaching developers to think in terms of background work, guarded execution, and system-level observability. Once that mental model sticks, model quality still matters, but product stickiness matters more.

Wall Street gets agent templates.

Anthropic's May 5 move was not a new chatbot. It was a pack of opinionated finance workflows, Microsoft 365 reach, and a services wrapper designed to get the work actually deployed.

This is the part of the market that looks boring until it compounds. Once a finance team has a preferred diligence stack, governed data access, and review flows wrapped around it, the switching cost is not "which model is better next quarter?" It is "do we want to re-platform our workflow?" That is a much stronger position.

OpenAI turns depth into a moat.

The most useful OpenAI post this week was not about a model. It was about behavior: what frontier firms actually do differently once AI stops being just a chat tab and starts becoming part of the operating system for work.

The read here is straightforward. OpenAI has enough consumer gravity to reduce rollout friction inside companies, enough developer depth to make Codex a real wedge, and now a clearer theory for how to monetize both attention and work. That is a harder business to attack than a pure model vendor.

Google bets on reach and speed.

Google's story is still easy to underrate because the launches feel less theatrical. But the underlying position keeps getting stronger: Gemini 3 Flash is rolling into mass distribution, the API is already processing over 1 trillion tokens per day, and the app reach is enormous before I/O even starts.

• Gemini 3 Flash is rolling into the Gemini app, AI Mode in Search, AI Studio, Gemini CLI, Antigravity, Vertex AI, and Gemini Enterprise.
• Gemini 3 Flash is explicitly framed as pro-grade reasoning at flash-level latency, which is another way of saying Google wants the default "good enough and everywhere" seat.
• Google I/O 2026 on May 19 and 20 is now the next major narrative checkpoint, with Google already warming up the ecosystem around Gemini-built experiences.

If Anthropic is trying to own premium workflow quality and OpenAI is trying to own the operating layer for work, Google is trying to make Gemini feel native to every context where people already are. That is a different, but very serious, way to win.

ServiceNow opens the governed runtime.

The interesting MCP story this week was not another hobby server. It was ServiceNow turning its system of action into a generally available MCP surface, then wrapping that surface in identity, permissions, audit trails, and enterprise metering.

The sharpest implication is that MCP's enterprise future looks less like a messy open registry and more like a layered market: open protocol on top, governed registries and runtime control underneath. The protocol won mindshare. The runtime fight is just starting.

What to watch before next Friday.

This issue closed with the market more clearly segmented than it looked a week ago. The next turn is likely to be less about another isolated launch and more about whether distribution, governance, or premium workflow adoption sets the story heading into mid-May.

One more practical note: the coverage window here is May 2, 2026 through May 8, 2026. The most unstable part of the next week is not model release timing. It is whether usage, governance, and distribution numbers keep moving fast enough to redraw how people rank the three frontier providers.

That is the briefing. See you next Friday.

Built from the May 8 roundup, then tightened with current official sources from Anthropic, OpenAI, Google, ServiceNow, and Colorado's legislature. Semantic-first article in reading mode, deck-style pacing in present mode.