Nvidia’s $3bn boost for meta’s agents as bloomberg and virtana launch safe AI tools.

In a sign of AI agents tackling niche professional tasks, Bloomberg on 28 October introduced BQuant Assistant, a “domain-grounded” AI coding assistant embedded in its financial analytics platform ([1]). The assistant can answer users’ questions, generate Python code and advanced financial queries, and even execute code for data analysis, but only with the user’s explicit approval ([2]). BQuant Assistant is fine-tuned on Bloomberg’s vast financial data, drawing on 22,000 data fields and example models to provide relevant support for analysts and developers working in the Bloomberg Terminal environment ([3]) ([4]). By bringing AI-driven coding help directly into a secure sandbox, Bloomberg aims to save its clients time on repetitive programming without compromising the quality or control of their models. For firms, it’s an example of using AI agents to speed up complex workflows in a governed way, limiting the agent’s autonomy to reviewing documentation, suggesting code and automating steps only under human supervision.

IT operations are getting their own agent-driven upgrade. On 28 October, Silicon Valley firm Virtana launched an "Agentic Observability" platform that lets its AI agents recruit other companies’ agents to help resolve technology incidents ([1]) ([2]). By using open standards, including a new Agent2Agent protocol, a Virtana monitoring agent that detects a problem can automatically call in specialised agents from third-party IT service, cloud or security tools to collaborate on diagnosis and remediation ([3]) ([4]). The company says that what used to take human-equipped IT war rooms hours can now be handled by a team of agents “in milliseconds”, for incidents they have been trained and authorised to handle ([5]). Enterprises can configure exactly which outside agents are allowed to join and require engineers’ sign-off for certain actions, with every step logged for audit ([6]). For executives, it’s a glimpse of how autonomous software could slash downtime, and a reminder to set clear rules on where agents can operate and when humans must still step in.

A new official survey underscores that most organisations are embracing AI agents, but keeping tight control over them. The Bank of England and UK Financial Conduct Authority published their latest joint AI survey on 26 October, finding that 80% of responding financial firms already have AI agents in live use ([1]). However, only 2% of these companies allow their agents what regulators call “broad autonomy”, meaning agents can make and execute decisions independently ([2]). The vast majority of deployed agents are limited to decision support: they can recommend an action but a human must approve or carry it out ([3]). Businesses are especially cautious about fully autonomous agents in critical areas like client interactions, payments and trading, with under 10% of firms even planning to test agents in those roles in the next year ([4]). Transparent oversight of third-party AI systems is another concern: only about a quarter of firms say they have deep knowledge of how their key external AI tools work ([5]). The findings reinforce Lumo’s guidance that AI agents deliver the most value in well-bounded tasks where human supervisors own the outcomes, and that building trust for greater autonomy will take time, rigorous testing and strong controls.