The End of the AI Buffet

Because autonomous agents were consuming $1,000-$5,000/day of compute through subsidized subscriptions, bypassing prompt caching that reduces costs by 92%. It was about closing an exploit significantly more expensive to serve than same usage through Anthropic's own tools. [Answer]

Anthropic generates $30B with ~5% of OpenAI's user base by targeting enterprise deals. 85% of revenue comes from API and enterprise deployments, just 15% from consumer subscriptions. [Answer]

Models converge fast. Claude 3.5 Sonnet and Gemini matched GPT-4 within months. Open-source models like Llama 3 are good enough for many use cases. Foundation labs facilitate portability by building routing layers. [Answer]

Anthropic is building a single conversational interface where work itself happens - Claude Code for coding, Claude Design for visual work, Co-Work for business collaboration, creating an ecosystem hard to leave. [Answer]

First subsidize to attract users and capture market share, then tighten terms once users are hooked to extract profit. Same playbook as Amazon, Uber, Facebook - predates Silicon Valley by a century. [Answer]

Almost none - codebase lives in your repo, CLAUDE.md files are portable markdown. But it creates developer affection which creates procurement stickiness. [Answer]

Deep lock-in - shared projects, conversation history, company connectors, SSO, role-based access, audit logging all live inside Anthropic's application. Migration is ERP-scale. [Answer]

Contract lock-in (spend commitment duration), data/context lock-in (where org history lives), model lock-in (ease of swapping underlying AI model). [Answer]

Start by paying for Opus, prove value, move fast. Then evaluate migrating high-volume workloads to fine-tuned open models on own infrastructure. Prove value first, then optimize unit economics. [Answer]

Timing tells the story - Anthropic isn't just protecting margins, it's building a platform. The restriction was partly strategic, but safety concerns about unconstrained agents were also credible. [Answer]

More like Microsoft - bundling applications for enterprise buyers with consolidated billing. Google's experimental approach kills beloved but insufficiently popular products. Market will determine which wins. [Answer]

The era of infinite subsidized compute is over. Most dangerous: high data/context lock-in with unpredictable pricing. Negotiate contracts carefully, pay attention to where institutional knowledge lives. Plan accordingly. [Answer]