5 must-reads in AI this week

Anthropic's billion-dollar week

Anthropic closed financing at a $965B valuation — officially passing OpenAI as the world's most valuable private AI company — and confidentially filed for an IPO. The engine behind the surge is Claude Code, its AI coding assistant, which has quietly become one of the fastest-growing products in the category. The takeaway: AI coding tools are no longer a niche — they're the center of gravity, and the revenue, for the whole industry.

The agent land-grab is on

The market has stopped asking "are AI agents real?" and started asking "which part of my company gets agentized first?" The proof is in the M&A: Asana bought StackAI, Salesforce bought Contentful, Coupa bought Rossum, and Vertice bought Vendr — all to acquire capabilities their agents need to act, not just advise. When incumbents start buying instead of building, it means the window to ship is now.

Four big models, one month

Four major model storylines are landing in the same few weeks: Google's Gemini 3.5 Pro, Anthropic's Claude Mythos 1, a rumored Sonnet 4.8, and xAI's long-delayed Grok 5. Pricing is settling too — GPT-5.5 at $1.50/$9, Claude Opus 4.8 at $5/$25, Gemini 3.1 Pro at $2/$12 per million tokens. For builders, capability is no longer the bottleneck — choosing the right model for the job and the budget is.

Compute is the real war

Behind the model headlines is the actual battlefield: compute. xAI's filings revealed it is renting 300 megawatts to Anthropic for $1.25B per month through 2029, while Amazon's custom AI-chip business has crossed a $20B annual run rate, growing 100%+ year over year. Models are downstream of GPUs — whoever controls compute controls the roadmap, and the pricing you'll pay.

MCP becomes the standard

The connective tissue for agents is consolidating fast. Google shipped Managed Agents in the Gemini API at I/O, Alteryx launched an MCP Server, and AWS, Microsoft, IBM and Databricks now all describe agents the same way: goals, memory, planning, tool use, autonomy. The Model Context Protocol is becoming the default way enterprises wire agents into real systems. This is exactly the layer we build on — and where the next decade of enterprise software gets written.

Building with AI?

We turn this stuff into shipped software — agents, MCP tools, automation, and apps.