Daily AI Brief

Salesforce and Anthropic announced an expanded partnership aimed at bringing Claude models to customers in regulated and data-sensitive sectors via Salesforce's Agentforce platform and Slack. The move emphasises guardrails, auditability and deeper vertical tooling (e.g., financial services, healthcare). It also underscores a broader pivot from generic chatbots to role-specific agents embedded in enterprise systems.

Reuters reports Salesforce has broadened access to OpenAI's latest models within the new Agentforce 360, enabling users to create agents, analyse customer data and even run Salesforce functions from within chat surfaces. Together with Anthropic, it positions Salesforce as a front door for enterprise AI across multiple providers.

Citi told Reuters that AI tooling has freed up around 100,000 developer hours weekly, largely through code assistance and automation. The headline figure illustrates how major banks are operationalising AI beyond pilots, with measurable productivity returns and stronger internal demand signals.

WPP and Google announced an expanded, multi-year collaboration that blends Gemini, Google Cloud and ad-tech integrations to accelerate AI-assisted creative development, media optimisation and measurement across WPP's global client base. Expect faster creative iteration loops and more AI-native measurement in large campaign stacks.

AI-exposed chipmakers moved on shifting U.S.-China trade rhetoric and competitive news. Barron's flagged a pre-market dip in Nvidia alongside moves in AMD and Broadcom; the tape remains sensitive to geopolitics and hyperscaler capex linked to 10-gigawatt-scale accelerator deals.

Eric Schmidt argued there is "evidence" AI models could be repurposed for serious harm if misused or compromised, renewing calls for robust security, monitoring and red-team practices in advanced model deployments. His remarks feed into ongoing policy debate on mandatory safeguards for powerful frontier systems.

While not from the last 24 hours, Google DeepMind's Gemini 2.5 "Computer Use" model—released last week—continues to shape product roadmaps by enabling agents that can operate web and mobile UIs reliably with lower latency. It outperformed peers on web/mobile control benchmarks and is now appearing in developer workflows.

The latest Salesforce-Anthropic expansion strengthens Claude's enterprise distribution, especially in regulated segments where deployment assurances matter. It follows a stream of recent enterprise-scale rollouts and partnerships designed to broaden Claude's footprint beyond direct API consumption.

Salesforce's Agentforce 360 integrations with OpenAI and Anthropic push agentic patterns (task-oriented assistants with memory, tools and governance) into mainstream enterprise stacks. Expect more pre-built vertical agents and compliance artefacts to land in the coming weeks.

The Citi example offers a concrete benchmark for internal AI ROI, focusing on developer time saved. For leaders still wrestling with business cases, consider tracking "hours returned" alongside quality metrics (bugs reduced, incidents avoided) as early KPIs.

"AI won't replace bankers—bankers using AI will."

Paraphrasing a common industry refrain, sharpened today by Citi's 100,000-hours-per-week stat. (Use for internal rallying, not external marketing.)

Why it matters: Practical UI-control for agents—reliable web/mobile action execution with lower latency—unlocks real workflow automation beyond chat. Useful for RPA-adjacent tasks, QA, or wrapping internal tools.