Gartner says free-to-run AI models can handle most routine business tasks

Analyst firm Gartner says most of the work AI agents do for businesses is routine, and cheaper open models can handle it as well as the priciest ones. In a new paper, two Gartner analysts argue that nine in ten steps in an agent workflow do not need a top-tier model. The Register reported on the findings.

What Gartner found

An AI agent is software that carries out a series of tasks for you. Gartner says typical workflows include hundreds of routine steps, such as sorting data, calling other tools, passing information along, checking it and matching records. For every one step that needs advanced reasoning, the analysts say open-weight models can handle hundreds of routine ones. Open-weight models are AI models whose core files are shared publicly, so a company can run them itself.

The savings, and the catch

Gartner says an open model costs a quarter as much as the cheapest closed model that scores the same. Against a top-tier model used on every step, it costs as little as a twentieth. Those are best-case figures. The analysts admit open models can be wordier and retry more often. That brings the real saving per task closer to five or six times.

Control matters more than cost

Gartner treats the savings as a bonus. Its bigger point is control. The analysts note that years of workflow history are a large part of what a buyer pays for when acquiring a software company. Every routine step sent to a top-tier model's online service moves a slice of that history outside your own walls. Open models let a company keep it in-house.

Where the value sits

Gartner says most models are already commodities, meaning one is much like another. The valuable part is the harness, the tool that runs and manages agents. A good harness sends each job to the right model without users noticing. It also saves money, sets strong guardrails and manages groups of agents. Weaker harnesses leave users with more work.

What happens next

Gartner predicts that by 2030, 90% of business agent workflow steps will run on open-weight models. It expects companies to favor models built for one field, which work best when a few stable steps run millions of times a month. Top-tier models keep a role in research and development, the paper suggests. Open models then handle day-to-day running. This is a forecast, not a measured result.

What to do

• Ask your AI vendor which tasks truly need a top-tier model.
• Ask whether your data has to leave your control for routine steps.
• Ask how the product chooses which model handles each job.