Banks Move to Curb Soaring AI Agent Costs
McKinsey reports banks worldwide are becoming more selective about where they deploy agentic AI as token spending climbs.
Updated on Nov. 10, 2026 in Banking
Live Poll
Would you let an AI agent make a financial decision for you?
Banks worldwide are trying to curb rising AI agent costs, according to McKinsey, as spending on AI tokens could reach 20 percent of total bank labor costs. Some institutions have shifted responsibility for AI-related spending to new teams called AI FinOps.
Why it matters
Tokens add up quickly when AI agents communicate with each other, and the routing models banks first used to control costs are ineffective in that setting. Banks are therefore concentrating agentic AI on areas with the strongest expected returns.
McKinsey's survey of 46,000 retail banking customers found that 72 percent of AI users would let a banking agent carry out one financial task, while token spending could reach 20 percent of total bank labor costs. Whether banks can keep those costs in check as agent-to-agent traffic grows remains unresolved.
The players
McKinsey & Company
McKinsey & Company is a global management consulting firm that publishes widely cited research on banking, technology and financial services.
Meta
Meta is the technology company behind Facebook, Instagram and WhatsApp, and developer of the Muse AI agent.
The details
Initial cost controls directed employees to different AI models based on query complexity, but those routing approaches break down when agents communicate with each other. Most banks currently block external AI agents from authorizing transactions within their systems, and Meta's Muse agent was downloaded 5 million times in the US during its first 22 days after launch.
Timeline
Meta's Muse agent recorded 5 million US downloads in the 22 days after its launch.
McKinsey expects AI agents to become widely used within three years.
Market Dynamics
The 5 to 10 years mobile internet took to reach broad adoption offers a benchmark against which McKinsey's three-year expectation for AI agents stands out as an accelerated timeline. AI agents could also pressure bank profits by shifting deposits into higher-yielding accounts, a dynamic that echoes earlier debates over digital banking.
Customers who use AI to compare banking services may soon be asked to let agents complete routine financial tasks on their behalf. Banks' decisions on which agent activity to permit will shape how much of that convenience reaches everyday account holders.
The takeaway
Readers using AI for everyday banking should check whether their bank permits external agents to authorize transactions before granting that access. Monitoring how institutions set approval limits, such as caps on autonomous approvals, can help customers judge the safeguards in place.
Further reading
For more on how institutions manage new digital costs, see Banking.
Live Poll
Would you let an AI agent make a financial decision for you?
