The Bloomberg Terminal Is Getting an AI Makeover, Like It or Not

For its famous intractability, the Bloomberg Terminal has long inspired devotion, bordering on obsession. Among traders, the ability to chart a path through the software’s dizzying scrolls of numbers and text to isolate far-flung information is the mark of a seasoned professional.

But as a greater mass of data is fed into the Terminal—not only earnings and asset prices, but weather forecasts, shipping logs, factory locations, consumer spending patterns, private loans, and so on—valuable information is being lost. “It has become more and more untenable,” says Shawn Edwards, chief technology officer at Bloomberg. “You miss things, or it takes too long.”

To try to remedy the problem, Bloomberg is testing a chatbot-style interface for the Terminal, ASKB (pronounced ask-bee), built atop a basket of different language models. The broad idea is to help finance professionals to condense labor-intensive tasks, and make it possible to test abstract investment theses against the data through natural language prompts.

As of publication, the ASKB beta is open to roughly a third of the software’s 375,000 users; Bloomberg has not specified a date for a full release.

WIRED spoke with Edwards at Bloomberg’s palatial London headquarters in early April. We discussed the impetus for revamping the Terminal, whether traditionalists might balk at the change, and Bloomberg’s attempts to iron out hallucinations.

The following conversation has been edited for length and clarity.

WIRED: Shawn, tell me about the rationale for this overhaul of the Terminal.

Shawn Edwards: For years, Bloomberg has kept adding to this comprehensive dataset that we have. Often, finding the right piece of data in the sea of information is the deciding factor in whether you’re successful or not. It has become more and more untenable: You miss things, or it takes too long.

The primary problem we’re solving with generative AI is helping users to find key insights and synthesize a view of the world around a particular idea.

The concept is that untapped alpha lurks somewhere in the data, and ASKB will help to surface it?

Yeah. The user gets to ask the high-level question—the thesis that’s in their head—instead of asking for particular data points. ‘How is the war in Iran and a change in oil prices going to affect my portfolio?’ That’s a big, big question with so many dimensions. Can we synthesize that answer in minutes?

In a scenario where everybody is able to wade through the tangle of data, what will separate mediocre traders from the very best ones?

These tools are not magical. They don’t make an average [employee] all of a sudden great. The difference will be your ideas.

In the hands of experts, it allows them to do better analysis, deeper research—to sift through 10 great ideas when they might have only had time for one. If you’re a mediocre analyst, they’ll be 10 mediocre ideas.

Bloomberg pitches ASKB as a form of agentic AI. On its face, it looks more like a chatbot interface than something that necessarily automates tasks. What is agentic about ASKB?

There are earnings that come out every quarter. My job as an analyst is to be prepared for what might come up in that earnings call. For each company I’m preparing for, I’m looking at how their price compares to their peers, searching through lots of documents, looking at their fundamentals, and on and on. During earnings season, I’m not sleeping.

With ASKB, I can create workflow templates. I can write a long query, and say, ‘Hey, here’s all the data I’m going to need. Give me a synopsis of the bull and bear cases, what the Street is saying, what the guidance is.’ Now, I want to schedule [the workflows] or trigger them when I see this or that condition in the world.

We’re essentially talking about automating the type of legwork that might fall to a junior analyst. Do you expect tools like ASKB to have a knock-on effect in the labor market?

I do think there’s a big question of how to educate, train, and mentor junior analysts coming through the workforce. At least for the next few years, you can’t just accept what some [AI] system says. You still have to have that rooted understanding of your craft.

I don’t know if the world has the answer.

In finance especially, you don’t want to be basing decisions on faulty information. What has been your approach to minimizing hallucination?

There’s a lot of work that goes into making AI behave itself. You have to build in validations at every single step.

We have a validation check that looks at the information content of a summary and checks whether those facts are fully contained in the paragraphs we synthesized from. There’s a semantic language check, because these models have a way of inverting things. We even have a check on our citations. These are simple examples. We take a very conservative approach.

But with what degree of confidence can you tell customers they can depend on the information spat out of this thing? Have you encountered any hallucinations, despite all of these checks and balances?

First of all, Bloomberg never gives a buy or sell signal to somebody. Bloomberg is there for them to make their own decisions, so they have to own their decisions. They always have.

We can never say it’s perfect. More of the problem is that we might not answer a question fully.

This is where transparency comes into play. I think these systems should be used to drive users to sources, not to hide them, or abstract them away.

Do you expect ASKB to become the primary method for interfacing with the Terminal, or is it more like an extension?

This will be the new Terminal. This will be the primary way most interactions are happening—definitely where they are starting.

I don’t think GUIs [graphical user interfaces] are going away. I should be able to pick up my mouse and do something. But by and large, people will start their analysis and workflows through ASKB.

The Terminal has dominated its niche. But lately, people have asked whether it might be possible to vibe code something with a similar functionality, at lower cost. Is this overhaul partly an attempt to fend off that threat?

The direct answer is no.

We’ve been trying to do this for years—trying to build things like this using machine learning models and other AI techniques. We actually had certain natural language interfaces on our command line. But it was brittle. You had to ask the Terminal in very specific ways. It just wasn’t able to do what we wanted. But finally, we have the technology to do it.

Vibe coding is great for certain things. But you don’t vibe-code your way through a mission-critical decision making system.

Among diehard Terminal users, set in their ways, do you expect there to be grumblings about the intrusion of AI?

We are designing for those power users. They should get even more out of the Terminal. They pride themselves in knowing the path through the Terminal, remembering which functions to run, knowing how to do things, and showing people how to do it. But that’s probably not the best way for them to use it.

I want to get away from the day where people have to have sticky notes on their screen, reminding them which function to use. Instead, you just ask.