Frontier AI Or Budget Models – Which Strategy Makes Sense For Your Business?

Corporate tech is splitting into two very different camps when it comes to AI spending. Shopify is going all in, with CEO Tobias Lütke making frontier model usage compulsory for engineers and weaving it into performance appraisals. At the same time, cost-conscious enterprises are reining in their spending, dropping pricey mega-models for smaller, task-focused tech that gets the job done for less.

Treating this as a strict either-or decision between premium and cheap models glosses over what’s happening. Shopify’s own engineering approach is more nuanced than the mandate suggests, and the companies switching to cheaper models aren’t simply cutting corners. What makes this split interesting is what it reveals about corporate value, and whether premium AI power is worth the heavy price tag.

Shopify’s Secret Weapon: Distillation

The Shopify story is often reported as “mandatory frontier AI,” but the reality is more architecturally interesting.

The company routes AI work through an internal proxy that switches between providers, including Anthropic, OpenAI and Google, based on task requirements, cost and availability. This prevents vendor lock-in and means Shopify can shift spend between providers without disrupting workflows.

More importantly, Shopify uses a Universal Distillation Platform that takes outputs from frontier models and distils them into smaller, fine-tuned models for specific production tasks. The savings speak for themselves, with costs dropping up to 30-fold on repetitive and high-volume jobs. Their fine-tuned model for Flow automation alone was 2.2 times faster and 68% cheaper than the mega-model it replaced.

The model here isn’t “frontier everywhere” but “frontier for reasoning and prototyping, distilled models for production at scale.” The mandate for engineers to use AI tools is about raising the floor on AI adoption. The cost optimisation happens at the infrastructure layer, out of sight of most individual contributors.

What Happens When You Ditch Frontier AI?

The cost-cutters are dropping premium AI for specific, high-volume tasks with clear rules and measurable outcomes. Address formatting, classification, document summaries and support ticket routing don’t require heavyweight models. Broad capabilities are wasted there, whereas a lean, fine-tuned model delivers equal or better accuracy at a fraction of the cost.

The catch is that this only works when teams can define the task tightly and measure outcomes in production. Shopify itself found that offline benchmark success didn’t automatically translate to online success when it implemented its distilled Flow model. The distance between a model performing well in testing and performing well in live production, with real users and edge cases, is where many cost-optimisation projects stall.

The shift is visible in how AI spending patterns are moving. Enterprise interest in specialised AI platforms and narrower, more controllable models has grown as the initial frontier API spend has proved harder to justify at scale. A 2x to 30x cost difference between frontier and distilled models makes the optimisation worth pursuing even at modest production volumes.

Where Companies Win Or Lose With AI

Whether this split creates a real advantage comes down to what that AI budget is paying for in the first place.

Stanford’s AI Index Report highlights productivity jumps of around 26% for engineers and 14% to 15% in customer support. However, those averages hide a key nuance. The gains are far higher for complex, open-ended tasks that need frontier reasoning. Budget models handle routine work as effectively.

The companies most likely to build a durable advantage are those that get both layers right: frontier models implemented where the reasoning complexity justifies the cost, with distilled or cheaper models handling the production volume where they don’t. The risk in the Shopify-style mandate approach is that it treats all AI tasks as high-stakes reasoning problems. Pure cost-cutting carries its own risk: underinvesting in exactly the tasks where frontier capability creates the most value.

The AI strategy question for most businesses in 2026 is less about picking a side in the frontier versus budget debate and more about building the infrastructure and discipline to use both intelligently. Shopify’s actual engineering approach, as opposed to the headline mandate, is probably the more instructive model. The companies getting the most from AI spending aren’t those spending the most. They’re the ones who know which tasks deserve premium capability and which don’t.