Bloomberg Terminal 正在以生成式 AI 进行介面改造，核心是名为 ASKB（发音 ask-bee）的聊天机器人式层，用自然语言把越来越庞杂的资料（财报、价格、天气、航运、工厂位置、消费、私募贷款等）快速聚合成可用洞见。Bloomberg CTO Shawn Edwards 认为资料量增加使传统搜寻「越来越难以为继」：使用者会漏掉资讯，或花太久时间找出关键片段。
截至发表时，ASKB beta 对约 375,000 名 Bloomberg Terminal 使用者中的约三分之一开放（约 125,000 人），尚未公布全面释出日期。Edwards 强调工具不会把普通员工「突然变得很厉害」；差异仍在想法本身，但专家可把研究效率从只来得及评估 1 个点子提升到可筛选 10 个点子（品质差者则是 10 个平庸点子）。
在「代理式 AI」层面，ASKB 允许分析师为每季（每年 4 次）财报季建立可重复的工作流程模板：一次写长查询，产生多公司、多文件的同业比较、基本面、牛熊摘要、华尔街共识与指引，并可排程或在条件触发时执行。为降低幻觉，Bloomberg 以保守策略在每一步加入验证：检查摘要事实是否完全出自所引用段落、做语义一致性检查以避免颠倒、并核对引文；同时不提供买卖讯号，主张透明地把使用者导回来源。Edwards 预期 ASKB 会成为主要互动起点但不淘汰 GUI，并认为「vibe coding」不适合任务关键决策系统。
Bloomberg is remaking the Bloomberg Terminal with generative AI, centered on ASKB (pronounced ask-bee), a chatbot-style layer that lets finance professionals use natural-language prompts to condense an expanding mix of data (earnings, prices, weather, shipping logs, factory locations, spending patterns, private loans, and more) into actionable insight. Bloomberg CTO Shawn Edwards argues that as data volume grows, the traditional hunt for the right datapoint has become “more and more untenable”: users miss things or it takes too long.
As of publication, the ASKB beta is available to about one third of the Terminal’s 375,000 users (about 125,000 people), and Bloomberg has not given a full-release date. Edwards says the tools are not magical and won’t make an average employee suddenly great; the separator remains the quality of ideas, though experts can use ASKB to examine 10 strong ideas in the time they previously had for 1 (and mediocre analysts will simply produce 10 mediocre ideas).
On “agentic AI,” ASKB is meant to automate repeatable workflow legwork around quarterly earnings (4 times per year): analysts can write long queries, generate peer comparisons, document synthesis, fundamentals, bull/bear summaries, Street narratives, and guidance, then schedule or trigger workflows on conditions. To reduce hallucinations, Bloomberg takes a conservative, step-by-step validation approach: checking whether summary facts are fully contained in synthesized source paragraphs, running semantic checks to catch inversions, and verifying citations; Bloomberg also avoids issuing buy/sell signals and emphasizes transparency that drives users back to sources. Edwards expects ASKB to become the primary starting point for most interactions while keeping GUIs, and he argues “vibe coding” is ill-suited to mission-critical decision systems.