Corporations shift to AI routing to cut costs

Companies are looking to limit AI spending by moving away from top-of-the-line models for simpler tasks. Glean CEO Arvind Jain reports around 95% of corporate AI usage currently relies on the most expensive models, even though many tasks can be handled by cheaper alternatives.

Routing allows directing complex queries to powerful models and routine operations to fast and inexpensive ones. Cognition CEO Scott Wu estimates this approach can yield five- or tenfold savings on common tasks. For example, Cisco faced $200 weekly AI spend per employee; with 90,000 employees, that amounts to $900 million annually.

AI developers are responding. Cognition introduced a productivity guarantee, promising to make up the difference if its Devin agent doesn't deliver expected savings.