BankingNewsAI Daily Brief ·
Mercury equips AI agents with corporate cards and hard spending controls.
Banking AI
Financial institutions & fintech technology
Mercury is giving AI agents corporate cards with hard spending controls
Mercury launched Spend, which issues AI agents their own payment credentials, budgets and audit trails rather than routing purchases through a human employee’s card. The product turns agent-driven purchasing into a controllable payments workflow with distinct permissions and transaction-level accountability.
Action
Set a policy now for agent-initiated payments: require separate credentials, pre-set limits, merchant controls, approval thresholds and immutable logs before exposing bank payment rails to autonomous agents.
Schwab integration brings agentic account opening into advisor workflows
Wealth-management AI platform Zeplyn integrated Agent Nexus with Schwab Advisor Center, enabling advisors to open Schwab Advisor Services accounts through an agentic workflow and automating meeting preparation. This is a live example of AI moving from drafting and research into a regulated customer-onboarding process.
Action
Prioritize account-opening automation where agents assemble and validate data, but retain explicit human authorization at identity, suitability, disclosure and account-submission checkpoints.
M&T Bank appoints a dedicated head of AI engineering
M&T Bank named Kalyana Bedhu as head of AI engineering, creating a visible senior technical owner for building and operating AI capabilities. The move signals that at least one regional US bank is treating AI engineering as a distinct production discipline rather than an innovation or analytics extension.
Action
Clarify whether your AI program has a single engineering accountable executive with authority over platform standards, model deployment, reliability and control integration.
General AI
Large language models & AI infrastructure
Microsoft has put its first in-house reasoning model, MAI-Thinking-1, into Foundry
Microsoft announced MAI-Thinking-1, its first reasoning model built from scratch, and made it available through Microsoft Foundry. The team is explicitly seeking feedback on tool use, signaling a model aimed at applied enterprise reasoning and agent workflows rather than a benchmark-only release.
Action
Ask your Microsoft account team when MAI-Thinking-1 will be available under your existing Foundry controls, data terms, and audit tooling—and include it in your next internal model evaluation.
OpenAI is selling higher-capacity ChatGPT Business seats for heavy enterprise users
OpenAI introduced Premium seats for ChatGPT Business that provide five times the usage of Standard seats and remove the five-hour usage limit. The offering formalizes a two-tier enterprise consumption model as high-volume users shift from occasional prompting toward sustained, agent-like work.
Action
Segment AI licenses by workflow intensity and build unit economics around power users; do not price enterprise AI as a uniform per-seat productivity tool.
OpenAI’s enterprise data shows leaders moving from chat to task-executing agents
OpenAI reports that enterprise adoption is shifting toward ChatGPT and Codex agents that complete work, with frontier adopters pulling ahead of laggards. The meaningful transition is from employee assistance to integration with operational systems and software-development workflows.
Action
Identify two or three bounded workflows where an agent can execute—not just advise—and fund the connectors, permissions, evaluation and exception handling needed for production use.