Enterprises Deploy Frontier AI Models as Strategic Advisors

Enterprises rethink how they deploy premium AI models

Companies are adopting a new approach to managing AI costs: using the most expensive frontier models only for strategic planning and high-level decision-making, while routing routine tasks to cheaper alternatives.

Ameya Kanitkar, CTO of San Francisco-based AI measurement platform Larridin, frames the strategy in practical terms. Just as businesses wouldn't hire their most expensive lawyer for routine paperwork, they shouldn't deploy premium models for simple tasks. Instead, Kanitkar told Business Insider in July, models like Anthropic's Claude 5 should create workflow roadmaps and provide strategic direction, while smaller models handle execution.

Michael Murphy, a partner at Sydney-based AI consultancy Adaptovate, echoes this view. Premium models should tackle complex challenges like strategizing, building initial app prototypes, or problems requiring sophisticated reasoning. Day-to-day work—transcribing meetings, checking facts, or drafting creative briefs—belongs with what Murphy calls "lightweight flashlight models."

Why it matters

This architectural shift reflects growing enterprise anxiety about AI return on investment. After a period of experimentation where some companies encouraged unlimited AI usage, executives are now demanding measurable value from their AI spending. The advisor-executor model offers a concrete framework for cost optimization without sacrificing capability where it counts most.

Industry leaders validate the approach

The consultant perspective aligns with thinking from technology executives. Coinbase CEO Brian Armstrong predicted in a June post on X that within 12 to 18 months, 80 percent of AI workloads would run on models that cost 99 percent less than today's premium options. Armstrong argued that the most powerful models should be reserved for what he termed "IQ maxxing"—scientific breakthroughs and agent orchestration.

The end of tokenmaxxing

This strategic shift marks a departure from tokenmaxxing, a trend where companies gave employees free rein to experiment with AI and burn tokens without constraint. Some organizations, including Duolingo, even made AI usage a performance metric.

Now companies are pursuing multiple cost-saving tactics beyond the advisor model. These include model routing systems that automatically select appropriate models for different tasks, and adoption of open-source Chinese models like Moonshot AI's Kimi K3 or Z.ai's GLM-5.2.

Startups capitalize on ROI concerns

The focus on AI efficiency has created a new market category: AI-routing companies that help developers select models and monitor spending. These startups are attracting significant investor interest. New York-based OpenRouter raised $113 million in May at a $1.3 billion valuation. Concentrate AI, an OpenRouter competitor, secured more than $5 million in funding, according to Business Insider.

Larridin, Kanitkar's company, provides visibility into employee AI tool usage and helps organizations improve ROI from their AI investments.

These details were first reported by Business Insider.