Bloomberg's chief technology officer, in his own words, has a job that's half about building the future and half about preventing an embarrassing future from ever coming to fruition. “Part of my job is to not deal with nonsense in the company,” Shawn Edwards tells the Observer, and he's not laughing at all. The nonsense he's talking about is the great wave of generative-AI hype that has swept through every boardroom with a Bloomberg terminal. His job is to find the fine line between what his engineers can dream of and what the people who actually trade bonds, screen credit and prepare earnings calls are trying in every way to do—not what they are doing today, but what they are trying to achieve. According to Edwards' world view, that difference is the planning principle behind ASKB, an AI system for conversations, AI Bloomberg has built directly into the terminal.
Ask Edwards for a use case, and he explains the old way of doing things. “Prior to ASKB, a user would have to go to different places in the Bloomberg terminal to look at company fundamentals, to look at what the street estimates are, to look at different performance measures, company KPIs, and a different place to look at print, peer analysis and other performance data for that quarter—to read a lot of news about the company and invariably to read a lot of company and sales documents.” Edward is silent. “And again then they will have to put all this information together.”
Starting in February, Bloomberg users can ask the system (ASKB) to aggregate information. “It knows where to download and it knows where to get all this information and it gives you detailed analysis to prepare.”
The machine does not replace the analyst, and Edwards is careful, almost insistent, on this point. “ASKB does not do all of the analyst's work, but it does 80 percent of the work of collecting and compiling this information. The workflow varies by desk. For equity analysts, event preparation and thesis monitoring. For credit analysts, liquidity analysis and bond evaluation. The common thread is a research problem.
Trust Is Not A Factor
If there's one fascination that emerges from Edwards' account of the past few years, it's the credibility of an engineering discipline tied to technology that was never monetized in the first place.
“My focus, the focus of the last few years for my team, and our core engineering team, has been how to build reliable AI for our customers to make important decisions,” he said. Among the principles that shaped ASKB was the refusal to let the model speak for itself. “We definitely don't want it to reveal the answer to its knowledge of the world,” explained Edwards. Instead, ASKB is guided and supported by Bloomberg's decades of proprietary data, risk analytics and price generators—what Edwards calls “sources of truth.”
Getting there means building validators into every step of the process—some checking facts in real time (“you didn't make it true, I can check that… sum it up, and I can look at all the facts and compare the two”), others finding subtle failures, like learning a distorted feeling. On top of that is a continuous testing framework, “automated and manual testing to make sure we're really moving forward, nothing is moving, and nothing is going wrong with our system.”
The last layer is transparent. ASKB points users to the source of the material—categories, in millions of documents, that have yielded insight. It tells users the question they made. Share the analysis call he made. Edwards is not shy about how well this can be overlooked from the outside. “They're underestimating the complexity of these different layers that have to work together to get something reliable,” Edwards said, about a handful of clients who have tried to replicate what Bloomberg is doing with off-the-shelf AI models. “It's not easy. It's actually very difficult to direct the AI to do it. It wants to help a lot, and sometimes it doesn't.”
Part of the difficulty lies beneath the model, in the unpleasant task of making the data sources speak to each other. “How do you connect the data? How do you balance all these different sources of information and balance the data model that you can join all the different data sources?” Sadly, Edwards says the people leading that work can't just be AI engineers. “Domain experts are increasingly building our systems. They're the ones who help guide the AI—they say, 'No, you're doing it wrong.'
Even a CTO Has a Learning Curve
For all the difficulties of engineering, Edwards admits something of a human wonder. The tools are harder to use than anyone initially expected—including him. “There was a lot of expectation at the beginning that it was natural to use, to have a conversation,” he said. “But I think we all learned that, in fact, there is a learning curve. You have to put energy and effort into learning how to be a good user of these tools, and the more you put in, the more you get out. That was probably a little overwhelming from the beginning.”
Bloomberg is building in more personalization to soften that curve, allowing users to tell the system what they like, their surroundings, their habits, so that “over time the system gets smarter about how the system will respond to your questions.” But Edwards is adamant that this is the first innings, and that it raises the really unresolved question in the AI industry about how much memory is too much. “There's a lot of research and a lot of tools coming out about memory—how to compress memory, how to use memory, how much past interaction do you want to use? Right now, we're very strict about not to retain some information, so it is a kind of balancing act.”
Bloomberg's current CEO, Vlad Kliatchko, started the same year Edwards did, and the two developed engineering together—a detail Edwards offers as evidence of a company you can “really find and stay for a long time, or leave quickly.”
The tradition follows Michael Bloomberg's open, deliberately flat environment, where “anyone can get a big view.” Edwards describes his role as protecting and balancing that instinct, creating scenarios for various conflicts between a technical expert, someone with an art degree, an ex-portfolio manager, and a practical person, all on the same whiteboard. Building ASKB required breaking that organizational chart, at least temporarily. Bloomberg's custom products are built by separate teams handling different functions, with UX partners designing individual screens. An AI system that accesses all domains in a terminal does not fit that model. “The surface of ASKB and the way it works is different from the way we build other systems,” said Edwards. “The way you work together, the way it's built, it's completely different… using our old architecture in this new way of building a brand didn't work. We had to change our thinking.”
Hiring Articulation Skills
Asked what he looks for in recruits, Edwards says nothing about the pieces. You hire for communication—the ability to take a complex concept and explain it on multiple levels, as a physicist might to a child, a college student, or a PhD student. This requires the ability to listen and be flexible with your ideas, especially in diverse groups where value is often not a new concept in itself but a discipline that forces it. “There is some research that says different teams work better with different backgrounds, not because there are new ideas, but because it makes each player work harder to express their ideas, so they think about the problem better.”
As for why talent wants to work at Bloomberg in the first place, Edwards points to the breadth of data—asset models that feed into weather data, research that powers document analysis, streaming prices in real time, because “finance is a world” and “many, many different aspects of the world that affect finance.” Edwards also notes the speed of Bloomberg's impact. “You can build a feature or a product or a capability and actually go see the customers who are using it, talk to them, get feedback from them. That's exciting.”
Pressed on what has shaped his thinking, Edwards cites the history of Bell Labs and its various glory days—not surprising, given how keen he is on teams that tend to clash. Another unexpected choice is Hermann Hesse's Steppenwolfa novel that taught him that people put themselves in a limited view of who they are, when in fact “we can take different forms and use different parts of our personality and our minds to grow, and we can release different abilities at the same time.” His career has developed almost at times of self-imposed discomfort. “Uncomfortable challenges,” he calls them.
What's next for AI at Bloomberg? Edward replied with something almost surprised. “We're just looking at what the artificial AI and our approach will do,” he said. “We still have a lot in our vision of what we can achieve. This technology has allowed us to dream bigger and face problems that we had dreams about but could not build. Now we can build them.”
