Companies Shift from AI Tokenmaxxing to Strategic Model Routing

From unlimited use to strategic allocation

AI-powered companies are abandoning the "use as much as possible" approach that dominated early 2026 in favor of deliberate model selection based on task complexity. Morgan Linton, CTO of AI startup Bold Metrics, exemplifies this shift by directing his 16 engineers to specific models twice weekly—Claude Fable on low settings for some teams, GPT-5.5 on high for others.

The change comes after organizations from Uber to Microsoft discovered their AI bills had ballooned under tokenmaxxing, the practice of encouraging maximum AI usage. Now, technical leaders are implementing what they call "token hygiene"—routing intellectually demanding work to expensive frontier models while delegating routine tasks to older, cheaper alternatives.

Why it matters

This strategic approach to model selection represents a maturation of enterprise AI adoption. Companies are learning that newer doesn't always mean necessary, and that thoughtful resource allocation can deliver comparable results at a fraction of the cost. The shift also signals that AI spending is coming under the same scrutiny as other technology budgets, forcing teams to justify their tool choices rather than defaulting to the latest release.

Practitioners develop switching strategies

Coinbase CEO Brian Armstrong articulated the emerging consensus in a June 7 post, predicting that 80% of workloads would run on models costing 99% less within 12-18 months, with only 20% requiring cutting-edge capabilities.

Tanvi Pisal, a Big Tech UX designer, learned this lesson after "wasting months of tokens" using Claude to brainstorm designs from scratch. She now creates mockups in Figma first, then uses Claude only to build functionality around completed designs. She also brainstorms with ChatGPT—free through her enterprise plan—before moving refined concepts to Claude for final documents.

Chris Maconi, cofounder of AI startup Hechura, said he tests lower-tier models before committing to premium options. When setting up his OpenClaw agent, he started with Gemini models before switching to Anthropic's Haiku. "I'm not afraid to go and try some of these lower-end models to see if they can provide the intelligence that we need," Maconi said.

Routing platforms automate the decision

For teams finding manual model selection exhausting, routing platforms are gaining traction. These tools automatically assign tasks to appropriate models based on complexity, sometimes including open-source options. Ramp's lead economist Ara Kharazian reported that 5% of firms now use routing platforms, up from 1% last year.

David Gilmore, who runs routing company Rayline, said many clients experience a "FOMO moment" with new models before their API bills arrive. His software intercepts requests and determines whether cheaper alternatives would suffice.

Spencer Yang, managing partner at BlockSpaceForce, suggested an additional strategy: asking a cheaper model whether a more expensive one is needed for a given task. "The models themselves are actually getting really good at assessing their own complexity," Yang said.

Maconi attributed continued reliance on premium models to inertia. "People don't want to do the hard work of understanding which models are good at which things," he said. "They just want to ride the hype train."

These details were first reported by Business Insider.