TD Bank deploys an "agentic AI" pilot to automate mortgage pre-adjudication

What changed: TD Bank published a May 21 announcement that its Layer 6 research team built an agentic AI model to automate pre-adjudication for mortgages and HELOCs, claiming early reductions in processing from ~15 hours to under three minutes per application in internal tests.

Why it matters: For lenders and fintechs this is a concrete example of agents moving into regulated workflows—agents can radically speed document classification, income checks, and summary memo generation, but the gains depend on good monitoring, data lineage, and governance.

Try/watch: If you operate in finance, run tightly scoped pilots with explicit human signoffs and the bank’s equivalent of a Trustworthy AI review; monitor model errors, consent checks, and disparate impact metrics before scaling.

Google’s Gemini Spark and agent features reach mainstream coverage — integrated, permissioned personal agents

What changed: Coverage on May 21 describes Google rolling out Gemini Spark (an agent that connects to Gmail, Calendar, Docs and external services) for testing and staged subscriber rollout, with Google saying high‑stakes actions will require explicit user permission.

Why it matters: For small businesses and operators, platform-integrated agents promise real time savings (scheduling, inbox triage, synthesis) because they can use existing context, but they also raise privacy and third-party permission questions when agents access email, payments or bookings.

Try/watch: Start with read-only integrations (drafting and summarization) before granting transactional permissions; keep a log of agent actions and consent prompts so you can audit any automated spending or commitments.