HBR's Case for Small AI: Buy Only the Intelligence a Task Requires
Synthesized by Helena Ross, Strategic Think-Tank Desk
- HBR frames AI strategy around the minimum intelligence needed to hit a required quality level on specific tasks.
- Companies default to the most powerful models and broadest platforms without checking fit against the task.
- Firms already contracted to OpenAI, Anthropic and other frontier labs may have better options for specific needs.
Harvard Business Review argues that AI strategy should start from the minimum model capability a task requires, not from the most powerful platform on offer.
«The Case for “Small AI”»
Summary and commentary by Aura-V. Facts, figures and opinions belong to the original.
AI spending has a default direction: toward the most powerful model and the broadest platform on offer. Harvard Business Review proposes starting from the opposite end, with the question of what minimum intelligence a company must buy so that AI accomplishes the tasks it needs at the required levels of quality.
The minimum-intelligence question
The summary's observation is behavioural. As companies invest in AI, it says, they often gravitate toward the most powerful models and broadest platforms available without considering whether those are the best fit for the tasks they need to accomplish. The proposed alternative is a floor rather than a ceiling: enough capability to clear a stated quality bar on a stated job.
Seven discrete tasks
Most businesses are not trying to solve arbitrarily complex problems, the summary argues. Each is a bounded job with an output, which is the condition that makes a minimum-intelligence calculation possible.
Contracts already signed
Many companies have contracted with frontier AI labs for this work — OpenAI, Anthropic and the other brand names of the AI boom. The summary argues that there may be better ways to meet those specific needs without triggering the risks and costs associated with frontier AI. It does not enumerate the risks or the costs.
Our read: the proposal is less a vendor argument than a procurement sequence — specify the task and the quality threshold, then buy capability. It gives B2B teams a defensible basis for declining the top tier, on the condition that they can state what quality a task actually requires.
HBR turns AI procurement into a specification problem: state the task and the quality bar, then buy the intelligence that clears it. The summary asserts a mismatch between frontier-scale purchases and bounded business tasks, but supplies neither prices nor risk detail to test that claim.
Interested in implementing this technology inside your company?
Our studio designs custom agent meshes, data pipelines, and outbound GTM systems for forward-looking enterprises.