Nathan Lambert hosts this episode of the AI Daily Brief, defending the controversial practice of token maxing - incentivizing employees to maximize AI token consumption. The show covers Google's pre-I/O announcements including Gemini Intelligence, the growing AI consulting wars between major labs, and Anthropic's legal sector expansion.

Lambert argues against critics who dismiss token maxing as wasteful, contending that experimentation is essential for the transition from assisted AI to agentic AI. He addresses the resurgence of 'AI bubble' narratives triggered by reports of employees gaming usage metrics, particularly at Amazon where staff allegedly automate unnecessary tasks to inflate scores.

Google's Pre-I/O AI Announcements Signal Major Event Ahead

Google unveiled Gemini Intelligence, an agentic suite for Android that transforms devices from operating systems into intelligent systems, rolling out to Google and Samsung devices over summer.

The Google Book combines Android and Chrome OS with built-in AI assistant featuring novel interactions like 'jiggle the mouse' to summon Gemini instead of learning hotkeys.

Google joins the orbital data center trend with SpaceX talks for space-based computing, following Anthropic's interest and NVIDIA posting job ads for orbital system architects.

Google Cloud plans to hire hundreds of forward-deployed engineers for AI consulting, directly competing with OpenAI and Anthropic's consulting initiatives.

The Token Maxing Controversy and Gaming Allegations

Financial Times reported Amazon employees using AI tools for unnecessary tasks to inflate usage scores, with managers tracking consumption creating 'perverse incentives.'

Meta's internal leaderboard ranks top 250 AI power users among 85,000 employees with titles like 'Session Immortal' and 'Token Legend.'

A viral Slack screenshot showed someone spending $600 on Anthropic being praised while $23 Uber Eats exceeded the $20 meal limit, highlighting perceived priorities.

Critics invoke Goodhart's Law: 'when a measure becomes a target, it ceases to be a good measure,' arguing token metrics create gaming behavior.

Defending Token Maxing as Essential Experimentation

The shift from assisted to agentic AI represents a fundamental change in work primitives, not just skills - 'managing agents is a new work primitive where there are no experts.'

Historical enterprise AI adoption studies from 2023-2025 consistently showed employees' biggest barrier was lack of time for experimentation alongside regular duties.

Lambert argues his billion tokens used last month had 'massive learning value' despite 0% direct financial gain, improving future token efficiency through experimentation.

Companies can easily identify gaming through traceable activity - 'Jim from Accounting goes from zero to a billion tokens' will prompt questions about outputs and learnings, not just rewards.

Anthropic Expands Legal AI with Claude for Legal Suite

Legal professionals became the most engaged Claude users among knowledge workers, prompting expansion with connectors to DocuSign, Trellis, and Thompson Reuters.

Claude for Legal includes 12 pre-built agents for commercial law, regulation monitoring, employment law, IP, and client management practice areas.

Cowork can now connect directly to Harvey AI, allowing legal professionals to use it as an 'agentic harness' for Harvey's legal knowledge base.

Anthropic's pattern involves rolling out branded packages of connectors and pre-built agents, contrasting with OpenAI's 'super app strategy' routing all knowledge workers to a single interface.