BT150 zeitgeist: Pick the best model for the job, think open models

Model routing and LLM portfolio management have moved to the front of enterprise priorities since it makes sense to mix and match open and frontier models, according to CxOs on Constellation Research's BT150 meetup. CxOs also talked about data center financing and enterprise adoption.

The BT150 meetup was held November 13. This Constellation Research CxO call operates under Chatham House rules so the takeaways aren't attributed.

• BT150 zeitgeist: SaaS pricing, AI agent looming questions, organizational design
• Meet the 2026-2027 BT150: Top Digital Transformation Leaders Driving Disruption and Innovation (full list)
• BT150 zeitgeist: Cloud capacity strains, AI agent governance, beware faux FDEs
• BT150 CxO zeitgeist: AI agent growing pains abound
• BT150 zeitgeist: It’s time to get practical about AI agents
• BT150 zeitgeist: In search of AI agent ROI

Here’s a look at the takeaways:

Usage of open models was a big theme. BT150 members favored matching models based on tasks, quality, cost, security, compliance and latency. Reserve frontier models for tasks that demand them and go cheap for routine tasks. Most expected frontier model companies to offer a wide range of models to optimize for price and performance. However, the expectation is that frontier labs won’t be able to undercut open models.

Model routing and specialization are central to AI adoption. Teams are segmenting workloads and routing sensitive or predictable tasks to in-house or smaller models. One participant summarized the principle as: “the least expensive model and architecture that reliably meet the business requirements.”

Focus on orchestration and intelligence layer. One BT150 member noted that his enterprise is focused on the orchestration and intelligent layer to classify tasks, route models, validate policies and add observability. You have to be model agnostic and focus on the right model for the job.

Purpose-built models deliver outperformance. The conversation pointed to transaction models for fraud and behavioral patterns, tabular models for credit scoring and churn, and time-series models for energy and demand forecasting.

Managing models is critical because they dominate AI costs. BT150 members noted that AI model costs have surged after being a relatively small percentage of the AI budget.

No consensus on AI reaching recursive self-improvement and what follows. Some participants saw existential-risk predictions as partly marketing; others stressed that increasingly capable systems could still cause harm without human-like cognition. The debate highlighted uncertainty around recursive self-improvement, model reasoning, and how to monitor systems effectively.

AI infrastructure growth is going to be constrained. Participants questioned how self-sustaining the current market is without circular investment, while also emphasizing debt exposure among some “neo clouds,” data-center opposition, and grid capacity.

Enterprise adoption is as much an organizational challenge as a technical one. The group discussed governance, accountability, interoperability, process redesign, and change management. As one participant put it, “We’re getting good at just feeding work to AI” but are still figuring out how organizations should redesign work around it.

• Open models gain traction in enterprises, tech CEOs say
• Evidence Shows Enterprises Use of Open Weight Models is Mainstream
• Open-Weight Models Are Gaining Ground in Enterprise AI
• Rightsizing open models may cut your AI inference spend
• Nvidia’s Nemotron hit enterprise AI inflection point