Bloomberg just embedded AI agents inside the Terminal. Not a separate product. Not a chatbot on the side. Agents that work in parallel across the same data universe that 325,000 financial professionals stare at all day. At thirty-two thousand dollars a year per seat, institutional knowledge just became executable code.
The Bloomberg Terminal has not fundamentally changed its interaction model in forty years. Users type commands. Screens return data. The human navigates, synthesizes, decides. The interface is dense, command-based, deliberately opaque to outsiders — a professional instrument optimized for speed in the hands of people who have spent years learning its grammar. At roughly thirty-two thousand dollars per seat per year, it is the most expensive software subscription in mainstream professional use. Three hundred and twenty-five thousand financial professionals use it daily.
This week, Bloomberg began rolling out ASKB — Ask Bloomberg — a conversational AI system now in beta that embeds agentic AI directly into the Terminal. Not alongside it. Inside it.
The Architecture
ASKB operates through a coordinated network of AI agents that work in parallel. A user asks a question in natural language — about a company, a sector, an investment thesis — and multiple agents simultaneously retrieve, interpret, and synthesize information from Bloomberg's entire content universe. The system draws from 5,000 original Bloomberg News stories per day, 1.1 million curated stories daily from external sources, sell-side and independent research from over 800 providers, and the full depth of Bloomberg Intelligence, BloombergNEF, and Bloomberg Economics.
The models are a combination of commercial and open-weight large language models, aligned with what Bloomberg calls its Responsible AI principles. Every response includes transparent attribution to the underlying research documents, news sources, and data points. Users can trace any claim back to its origin — a specific filing, a specific analyst report, a specific data series.
Shawn Edwards, Bloomberg's Chief Technology Officer, described the system as enabling users to 'ask detailed questions in conversational language and receive comprehensive answers synthesized from our extensive data, documents, news, research, and analytics.' Early beta feedback, he said, shows ASKB 'driving efficiency, improving discovery, and helping users surface actionable insights at speed.'
The technical detail that matters most is what the agents cannot see. Bloomberg's data ecosystem is closed. The agents do not browse the open internet. They do not hallucinate from training data. They operate exclusively within Bloomberg's proprietary content universe — a walled garden that happens to contain the most comprehensive financial data infrastructure ever assembled. The hallucination problem that plagues general-purpose AI systems is solved here not through model improvement but through environmental control. The agents can only look where the data is trustworthy.
The Workflow
The feature that reveals Bloomberg's real ambition is not the conversational interface. It is ASKB Workflows.
A user describes a multi-step research task — pre-earnings preparation for a specific company, post-earnings analysis across a peer group, meeting prep for a client covering a sector rotation thesis. ASKB assembles a structured output in minutes: the relevant filings, the consensus estimates, the recent analyst revisions, the news sentiment, the comparable valuations, the risk factors. The output can be saved as a reusable template, rerun across different securities or time periods, and shared across teams.
This is institutional knowledge becoming executable. A senior analyst who has spent twenty years developing a proprietary research methodology can now encode that methodology into a workflow template. A junior analyst can run it across a different universe of securities. A trading desk can standardize its pre-trade research process. The methodology is no longer locked inside a single person's head or described in a training manual that nobody reads. It is code — not programming code, but a structured description of analytical intent that agents can execute.
The output extends beyond the conversational interface. ASKB generates Bloomberg Query Language code that users can export to Excel, BQuant Desktop, or BQuant Enterprise for deeper quantitative analysis. It surfaces relevant Bloomberg community connections — analysts covering the same names, strategists with adjacent views. It is available on mobile through the Bloomberg Professional app on iOS, Android, and Apple Vision Pro.
At thirty-two thousand dollars per seat, Bloomberg is not selling AI. It is selling the codification of expertise at the most expensive professional interface in finance.
The Protocol
Bloomberg's technical infrastructure for ASKB adopted the Model Context Protocol — MCP — originally created by Anthropic as an open standard for connecting AI agents to external data sources and tools. Edwards described MCP as 'the way we should talk to the world.' Bloomberg's implementation is remote-first and multi-tenant, with middleware handling identity management, access control, and observability.
The adoption is structurally significant. Bloomberg is a founding member of the Agentic AI Foundation, a directed fund under the Linux Foundation established in December 2025. The other founding members include Anthropic, Block, OpenAI, Google, Microsoft, Amazon Web Services, and Cloudflare. MCP now has over 10,000 public servers and 97 million SDK downloads per month.
The tension is visible from the right angle. The United States government expelled Anthropic from all federal agencies in February 2026. Defense contractors began dropping Claude. The State Department switched to OpenAI. And in the same weeks, the single most important infrastructure provider in global finance adopted Anthropic's protocol as the standard for how its agents connect to the world. The protocol's creator is politically radioactive. The protocol itself is becoming the plumbing.
This is how standards work. They outlive the controversies of their creators. TCP/IP was a DARPA project. HTTP was a physics lab side project. MCP may end up the same way — the protocol that survived by being useful after its origin story stopped mattering.
What Changes
The Terminal is where information terminates in decisions. That is literally what the word means. For four decades, the terminal point was a human — a portfolio manager scanning screens, a credit analyst cross-referencing filings, a trader watching order flow. The information flowed through Bloomberg's infrastructure and arrived at a person who synthesized it into action.
ASKB does not remove the human from that terminal point. It multiplies the eyes. An analyst who previously spent forty-five minutes preparing for an earnings call can describe the preparation in natural language and receive a structured brief in minutes. The time saved is not idle time — it is time redirected toward the judgment calls that no agent can make: whether the CEO's tone on the call signals something the transcript won't capture, whether the consensus revision trajectory matters more than the consensus itself, whether the position sizing reflects the conviction or just the habit.
The deeper shift is distributional. Institutional knowledge has always been the most valuable and least transferable asset in finance. A senior analyst's pattern recognition — the instinct for which data points matter in which context, developed over decades — dies when that analyst retires. ASKB Workflows do not capture the instinct itself. But they capture its output: the specific sequence of analytical steps, the specific data sources consulted, the specific structure of the final product. The methodology becomes infrastructure. The judgment remains human.
David Easthope, an analyst at Coalition Greenwich, predicted that AI will 'unlock new insights, automate complex analyses, and drive efficiency' in 2026 investment research, with 'its full potential only just beginning to be realized.' That framing — efficiency, insights, potential — is the standard industry narrative. It is also insufficient.
What is actually happening is that the screen through which finance sees the world now has agents behind it. The Terminal was always the lens. The question was always who was looking through it. The answer used to be obvious. It is becoming less so.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)