brickster.ai

Brickster Newsletter

AI agents are software tools that can take several steps to complete a task, such as handling a support request or preparing a report. A new survey from G2, the business software review platform, shows why buying an agent is only the start of the budget conversation.

G2 surveyed 1,019 business software buyers in July 2026 and published the findings on October 14. These are self-reported results from the buyers surveyed, rather than a measure of all businesses. The 86% figure covers purchases in the previous 12 months; live use can be limited to a team or workflow.

The costs extend beyond software usage. G2 found that 31% of buyers said employees spend at least six hours a week overseeing agents. That includes reviews, approvals and corrections. G2 calls this the distrust tax. Just 27% said their organization keeps a central register of agents. For budget owners, a shared list is a practical place to start.

That makes Databricks’ October 15 access change relevant. Business users with “consumer access” can now use AI model services already shared with them without an extra admin setup step. Existing permissions still apply. That could increase use, so ask your platform team which users and services it affects.

Unity Gateway, Databricks’ service for managing AI access and usage, lets admins set monthly spending thresholds, send alerts, block further use or limit request volumes. The spending controls use estimated costs, so the final bill can exceed a threshold. They cover supported usage through the gateway. Including external model costs requires a separate option that is still in Beta; ask your platform team which charges are included.

There is also maintenance to plan for. On October 12, Databricks marked Knowledge Assistant and Supervisor Agent as legacy products, meaning they are no longer recommended for new agents. If your team uses either, ask the platform owner to review ongoing support and any future migration work before you finalize the budget.

Before you approve the 2027 AI budget, agree three things with your business and platform teams:

1. Give every agent an owner. Keep one list of what each agent does, who uses it and who is responsible for its cost.

2. Set limits before expanding access. Agree a monthly allowance, alerts and who can approve more spending. Check which charges the controls actually cover.

3. Measure cost and value together. Include software, infrastructure and human review. Track the cost per successful task, such as a correctly resolved support request, and check whether the time saved justifies the expense.

The budget question connects to a wider one: how will you run and support these services? Recent news brings changes in data security, databases and migration. Here is what buyers should watch.

To turn these checks into a budget discussion, start with two guides from our archive: where Databricks shows AI costs, and how its Genie products differ.