Anmol Verma, who spent several years in public markets before founding the AI wealth management platform Finn, has set out how investment technology moves from reactive tools to proactive agents. His argument: the promise of agentic finance is not that investors make more decisions, but that they bring more intelligence to every decision without being limited by how much information one person can track. For business, that reframes where the value of such systems sits.
Who stands behind the argument
Verma's position comes from time spent in public markets before he founded Finn, an AI wealth management platform. He describes the shift as a change in the role of financial technology itself: from products that wait for investors to direct them to systems able to understand enough context to determine next steps. The distinction matters because today's products already automate a considerable amount of work. Portfolios can be rebalanced, risk monitored and trades executed according to predefined rules, but those systems act within narrow workflows built around a specific event or instruction.
Agents are presented as a more flexible form of automation, one that reasons through changing circumstances rather than following a predefined path. Instead of executing a single rule, an agent can interpret changes across a portfolio, relate them to an objective and coordinate the steps required to respond. For retail and institutional investors alike, that could mean understanding how new information affects an investment thesis or portfolio, identifying what deserves attention and helping determine what should happen next. The unit of work changes from a triggered instruction to a chain of related steps.
What this means for business
For companies that adopt these systems, the practical consequence is a different division of labour between people and software. Verma expects adoption in phases: agents may first help investors understand what is happening, then recommend what to do, and eventually take on more of the work required to carry a decision through. Early tasks are contained — updating a model after earnings, monitoring developments against an investment thesis, identifying areas that warrant further research. As systems become more reliable, they could take on broader parts of the process, from proposing new areas of research to recommending portfolio changes.
Being proactive is not about detecting more signals, and this is where the limits sit. An agent that reacts to every market movement, company announcement or missed target would create more work, not less. It needs to judge which changes are relevant, how urgently they matter and when no action is warranted. That requires context: the same announcement can be market noise or evidence that an important assumption behind an investment has changed, depending on the portfolio, objective and time horizon. Agentic finance also raises the stakes, because a system can understand the objective and still make the wrong decision — misread a thesis, miss a risk or act on incomplete information. The more responsibility it takes on, the more confidence investors need in its judgment and its boundaries.
Verma's answer to that risk is a learning loop rather than a single deployment. Every interaction reveals which recommendations are acted on, which are ignored, where an investor overrides a decision and which suggestions lead to better outcomes. Over time those signals let the agent learn from the decisions it supports, not just the information it processes, so recommendations and actions can become more relevant as understanding of the investor deepens. The stated goal is not to automate every investment decision but to automate more of the work around a decision while keeping investors focused on the areas where judgment matters most.
The marker to watch is the phase boundary rather than the product launch. When agents move from explaining and recommending to carrying decisions through, the question for a business becomes who signs off on an action and where the override sits. Until vendors can show which tasks an agent completes end to end and how it reports a wrong call, adoption stays in the contained tasks Verma describes — model updates after earnings, thesis monitoring, research triage.
