The news
Allie Miller, CEO of Open Machine, outlined how frontier technology companies structure their AI spending. She told Bloomberg’s Romaine Bostick that these firms separate their work into two distinct efforts. One effort applies modest subscription-level budgets to raise baseline capabilities across standard departments. The second effort gives a smaller frontier team much larger budgets and wide latitude for experimentation.
Context
Miller’s comments address the practical problem of matching AI costs to measurable returns. Companies have moved past the initial phase of uniform high spending on every AI project. They now treat most internal functions with controlled, repeatable tools while reserving open-ended investment for a limited set of advanced initiatives. This split reflects pressure to show clear financial discipline without halting progress on the most ambitious work.
The pattern emerged after several years of broad experimentation in which many organizations applied similar levels of funding across departments. Early projects often carried high per-user costs with uncertain productivity gains. Over time, finance and procurement teams began demanding clearer separation between routine efficiency gains and true research-and-development bets. Miller’s description captures the current equilibrium at firms that continue to push model capabilities while keeping overall spend visible to investors and boards.
Details
Miller named finance, legal, marketing, and sales as the departments that receive the baseline treatment. These groups use established AI subscriptions to improve routine tasks without large additional outlays. The frontier team, by contrast, operates with significantly higher budgets and fewer constraints on what it can test. Miller presented the division as the current pattern at companies operating at the leading edge of AI development. No specific dollar figures or company names beyond the general category of frontier firms were provided in the interview.
The baseline layer focuses on tools already available through standard vendor contracts. These tools target document review, data summarization, campaign drafting, and basic code assistance. Because pricing is tied to seat counts or usage tiers, costs remain predictable month to month. The frontier layer, however, may involve custom model training, extended inference runs, or partnerships with research labs. Those activities carry variable and often higher unit costs that would be difficult to justify across an entire organization.
Miller did not discuss exact budget ratios or head-count allocations. She did emphasize that the frontier group remains deliberately small so that its spending does not dominate the overall AI line item. This keeps the experimental work visible to leadership while protecting the larger organization from open-ended cost growth.
Why it matters
The two-tier model Miller described gives companies a concrete way to control overall AI spend while still pursuing breakthroughs. Departments that handle day-to-day operations gain incremental productivity without the risk of runaway costs. The smaller frontier group keeps the possibility of step-change advances alive, but its limited size prevents the entire budget from drifting into unproven territory.
For software engineers and technical leaders, the approach signals that broad AI adoption inside established functions will likely stay tied to predictable subscription pricing. Experimental work, meanwhile, will remain concentrated in small, well-funded teams whose results are judged on a longer horizon. This structure reduces the chance that every team claims frontier-level resources and instead forces clearer prioritization.
Over time the pattern may also shape vendor pricing. Suppliers of baseline tools will compete on cost per seat, reliability, and integration depth. Vendors serving the frontier side will continue to sell high-margin capacity with the understanding that only a narrow set of customers will pay those rates. Procurement teams can therefore negotiate two separate contracts rather than a single blended agreement that hides the experimental spend inside general operations.
The model also affects career paths. Engineers interested in large-scale experimentation will likely need to join the designated frontier teams rather than expect every product group to fund open-ended research. Conversely, teams focused on finance, legal, marketing, and sales will optimize for steady gains from existing platforms. This division of labor may reduce internal friction over resource allocation while concentrating the highest-risk bets in a visible, accountable group.
---
Sources:
{"word_count": 682, "sources_used": 1}
No comments yet