BankingNewsAI Daily Brief ·
AT&T cuts AI-task costs 56% by routing employee work across open models.
Banking AI
Financial institutions & fintech technology
Blend reports early lending gains from its pre-underwriting agent
Blend says lenders using its Autopilot pre-underwriting agent achieved 10–15% higher pull-through rates and materially reduced processing time in early production. The agent handles work ahead of underwriting, targeting the document and data friction that delays loan decisions.
Action
Test whether pre-underwriting automation can improve conversion without weakening credit-policy controls, and require segmented results by product, borrower cohort and exception rate before scaling.
General AI
Large language models & AI infrastructure
AT&T is shifting most employee AI work to open models after cutting coding costs 56%
AT&T reports that 40% of employee AI usage already routes to open models and is targeting 60–70%. Its coding workloads cost 56% less with only a reported 2% quality decline, at a scale of 45 billion tokens per day. This is a concrete enterprise signal that hybrid model routing is becoming a cost-control architecture rather than an experiment.
Action
Pressure your AI platform and systems-integration vendors to show workload-level routing, quality controls, data-residency options, and unit-cost comparisons between frontier and open models.
Anthropic will let enterprises meet Claude retention requirements in their own cloud
Anthropic plans to revise its policy requiring customers of its most advanced models to retain data for 30 days, allowing enterprises to retain it in their own cloud environment rather than Anthropic’s. The adjustment responds to enterprise concerns and directly addresses a recurring obstacle for regulated deployments: control over retention and data location.
Action
Re-open Claude use cases previously blocked by data-retention requirements, but validate contractual retention, deletion, audit-log and subprocesser terms rather than relying on the policy change alone.
AT&T cut AI-task costs up to 56% by routing work across models
AT&T reportedly reduced costs for coding and other advanced AI tasks by as much as 56% using model-routing tools and open-source models, while performance declined only 2%. The result is a meaningful at-scale proof point for using a portfolio of models instead of sending every request to the most capable—and expensive—provider.
Action
Deploy a governed model-routing pilot for internal copilots: classify tasks by sensitivity and quality requirement, measure unit cost and accuracy, and reserve frontier models for tasks that demonstrably need them.