To explore the implications of agentic AI for investment banking, of course I turned to ChatGPT. I asked it to conduct a conversation with me—David Brunner, Founder and CEO of ModuleQ—about how this emerging class of AI systems is reshaping knowledge work, and how ModuleQ is applying agentic AI to enhance workflows for bankers. Drawing on themes from the recent Harvard Business Review article, “What Is Agentic AI, and How Will It Change Work?”, we discussed how AI agents can bring domain expertise to bear on complex tasks, reduce information overload, and ultimately help bankers focus more of their time on high-value, strategic work. ChatGPT also served admirably as editor, revising my answers for brevity and clarity!
David: Agentic AI refers to artificial intelligence systems that can take initiative—they don’t just respond to prompts but act proactively to accomplish tasks. Definitions vary slightly across major players like Microsoft, Google, IBM, and NVIDIA, but the common thread is agency: the ability for AI to make decisions and execute actions on its own.
What makes agentic AI so exciting right now is the maturity of large language models. These models have developed a kind of synthetic common sense—the ability to interpret complex information, plan a sequence of actions, and choose the right tools to achieve a goal. Historically, this kind of reasoning was well beyond AI’s capabilities.
What’s changed is that we can now pair LLMs with scaffolding frameworks and software tools that allow them not just to plan, but to actually act. That means AI can complete complex workflows that used to require human effort or demanded intensive, brittle automation. In short, agentic AI lets organizations get meaningful work done with less human intervention—and with far greater flexibility.
David: That article – “What Is Agentic AI, and How Will It Change Work?” – makes an important point: specialization is the key to unlocking the power of agentic AI. Large language models are impressively general, but that generality is also a weakness when it comes to complex professional work. To make agentic AI effective, we have to constrain its scope—focus it on specific workflows, tailor the tools it can use, and tightly curate the information it draws from. That’s when LLMs become not just conversational partners, but powerful task performers.
In investment banking, there’s a wide range of highly specialized, high-effort tasks that are perfect candidates for agentic AI. Think of researching a company or industry, preparing a pitch book, running a valuation model, or identifying relevant comparables. These tasks consume huge amounts of analyst time, but they follow recognizable patterns and require access to structured tools and trusted information sources.
When we embed agentic AI in these workflows—with the right domain-specific tools and data access—it can collaborate with bankers to generate first drafts, surface key insights, and iteratively refine outputs. Crucially, the AI is operating within a tailored environment that reflects the task at hand. That level of specialization is what enables the AI to actually deliver value in a field as complex and bespoke as investment banking.
David: Traditional automation and rule-based systems have had very limited success in investment banking, and that’s no accident. The work is inherently bespoke—every deal, every client engagement has its own context, nuance, and strategic goals. That’s why investment banking remains such a human-intensive business.
What makes agentic AI different is its ability to adapt to these nuances in real time. Thanks to the flexibility of large language models, agentic systems can process unstructured input, understand the specific context of a client or deal, and take appropriate actions using a variety of tools. That’s a major leap from rigid automation.
For example, instead of applying a one-size-fits-all template, an agentic AI can help draft a pitch tailored to a particular sector or client strategy, and then refine it based on banker feedback. It can reason through context the way a junior analyst might—only faster and at scale. That’s what opens the door to true augmentation of knowledge work, not just task automation.
David: The concept of “tools” in agentic AI can sound abstract, but it’s very concrete in practice. When we say an AI agent uses tools, we mean it can call specific software functions—APIs that perform domain-relevant actions like retrieving deal data, pulling earnings estimates, or calculating valuation multiples.
The key is that these tools help the AI focus. On their own, language models can be surprisingly unfocused or probabilistic in how they reason. But when you give them access to a curated set of tools and knowledge sources, they become dramatically more capable.
In our work at ModuleQ, we’ve seen this firsthand. For instance, when we gave an agentic system access to a tool for filtering M&A transactions by key characteristics, it was able to zero in on the most relevant deals with far more accuracy than a general-purpose LLM—even though both had access to the same data. The difference was the tool, which acted as a cognitive enhancer, helping the AI structure its reasoning around what mattered.
That’s why specialization is so critical—not just in language and knowledge, but in tooling. It gives agentic AI the precision it needs to be truly useful in a high-stakes environment like investment banking.
David: At ModuleQ, we’ve focused on a very specific challenge that bankers face: information overload. In fast-paced, relationship-driven environments like investment banking, the key to success is being timely, relevant, and insightful with clients. But staying on top of all the information that might support that is nearly impossible—there’s simply too much of it, and too little time.
That’s where agentic AI comes in. Our system continuously works in the background to surface the right information for each banker at the right moment. It scans research, news, deal data, and other sources, and matches that information to each banker’s relationships, coverage areas, and current workflows. Then it delivers that intelligence proactively—via daily briefings, meeting prep, and real-time alerts—so bankers are better prepared to engage clients strategically.
This isn’t just summarization or search. It’s an agentic system that’s context-aware and personally tuned to each user, so it can anticipate their needs and reduce the cognitive load of finding the signal in all the noise. Ultimately, it empowers bankers to spend more time delivering value to clients and less time wrangling information.
David: Hallucination is a real challenge with language models, especially when they’re used to generate content. But at ModuleQ, we’ve approached this differently: rather than asking LLMs to generate new information, we use them to curate and prioritize real content—news, research, and structured data that already exists.
Because the content itself comes from trusted sources, the risk of hallucination is eliminated. The LLM is used to decide what’s most relevant, not what’s true. And even that decision-making can be augmented with quality controls—like comparing outputs to human judgments or combining multiple models to cross-check selections.
In many cases, we find that LLMs are actually more consistent than humans when it comes to curating information under time pressure. They’re less prone to fatigue, and they can make contextually smart decisions about which insights to surface, especially when they've been personalized to the user’s interests and workflow.
So the focus isn’t on replacing human judgment, but on enhancing it—using agentic AI to cut through the noise and deliver high-signal content without making up facts.
David: Personalization is central to how ModuleQ delivers value. Our system learns from bankers’ feedback—explicit and implicit—to continuously refine what it surfaces. For example, when a banker indicates that a piece of content was helpful—or ignores something repeatedly—the system takes that signal into account. Over time, it builds a model of each user’s interests, relationships, and preferences.
What’s powerful here is that this learning happens at the individual level. Two bankers covering the same sector might have very different styles or strategic priorities. Our agentic AI doesn’t try to generalize across the organization; it tunes itself to each user’s unique context.
This kind of feedback loop is what transforms a useful tool into a trusted assistant. After just a few weeks of interaction, bankers start to feel that the system “gets” them—it starts surfacing content that’s sharper, more relevant, and more aligned with their goals. That’s the kind of value that earns trust and drives adoption in high-performance environments.
David: If you zoom out and look at how investment banking is structured, it’s a classic pyramid: senior bankers drive client strategy, while junior bankers do the heavy lifting—preparing pitch decks, building models, compiling comps, and so on. Much of that work is bespoke, but it’s also repetitive and time-consuming.
Agentic AI has the potential to change that equation. Instead of automating entire jobs, it can automate pieces of high-value workflows—supporting tasks like pitch creation, deal analysis, valuation modeling, and document preparation. That means more time spent thinking strategically and creatively, and less time spent wrangling information.
In the long run, I think we’ll see specialized AI agents supporting each stage of the deal process. The result won’t just be greater efficiency; it will be a shift in how bankers deliver value. With the burden of mechanical work lifted, they can focus more on exploring opportunities, advising clients, and bringing judgment and insight to the table.
And as these tools become more powerful and accessible, the economics of investment banking may start to shift. Lower costs could mean broader access to advisory services, opening the door for a more agile and dynamic flow of capital across the economy.