The Sequence Radar #869: Last Week in AI: The Token Becomes the Unit of Account — Opus 4.8, OpenRouter, Cognition, Snowflake, and a papal warning
Summary
The AI industry has transitioned from speculative benchmarks to tangible revenue, with the "token" emerging as the primary unit of account. Anthropic's Claude Opus 4.8, described as a "modest but tangible improvement," now features effort control, dynamic workflows for parallel sub-agents, and enhanced honesty, while the company projects \$10.9B in Q2 revenue and closed a \$65B funding round. Concurrently, OpenRouter raised \$113M at a \$1.3B valuation, seeing its weekly token throughput surge fivefold to 25T. Cognition, developer of the AI software engineer Devin, secured \$1B at a \$26B valuation, with Devin now writing 89% of its internal code and run-rate revenue reaching \$492M. Snowflake further solidified this trend with a \$6B AWS compute deal and the acquisition of Natoma, signaling a data layer reorientation towards agent consumption. This economic shift coincides with Pope Leo XIV's encyclical, Magnifica Humanitas, which warns against AI's quiet disintermediation of human decision-making, linking the economic "meter" to the ethical concern of delegated judgment.
Key takeaway
For Directors of AI/ML evaluating new agentic system deployments or investment strategies, recognize that the "token" now dictates both economic models and the extent of human judgment delegation. Your focus should shift to understanding token consumption rates and the governance mechanisms within AI systems. Prioritize solutions like Claude Opus 4.8 that offer explicit effort control and self-checking capabilities, ensuring you can manage compute costs while maintaining necessary human oversight and mitigating the risks of quiet disintermediation as AI autonomy increases.
Key insights
The token has become the primary unit of account, quantifying both AI's economic value and the delegation of human judgment.
Principles
- AI development labs can achieve operational profitability, challenging prior assumptions.
- Technology is not neutral; it inherits the incentives of its creators and funders.
- The "token" serves as a dual metric for economic activity and human agency transfer.
Method
Claude Opus 4.8's dynamic workflows enable models to plan large tasks, deploy parallel sub-agents, verify outputs, and accept live message array edits mid-run for steering long jobs.
In practice
- Employ explicit effort control in agentic systems to balance compute and quality.
- Leverage multi-model routing services to optimize inference costs across diverse models.
- Train AI agents to surface uncertainty and self-flag code flaws for robust unattended operation.
Topics
- Agentic AI
- AI Economics
- Large Language Models
- Multimodal Embeddings
- AI Inference Routing
- AI Ethics
Best for: Entrepreneur, CTO, VP of Engineering/Data, Investor, Director of AI/ML, Tech Journalist
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by TheSequence.