Agentic AI systems are being marketed as a way for financial advisors to offload complex multi-step workflows with minimal human involvement. But early adopters say the technology’s real value today depends less on autonomy and more on strict governance, oversight and clear limits.
Processing Content
Unlike earlier AI tools that respond to single prompts, agentic AI systems are designed to execute end-to-end workflows on their own. These systems can operate across multiple platforms to pull data, monitor email and draft responses and make contextual operational decisions, often autonomously.
Proponents say agentic AI has the potential to dramatically reduce the time advisors spend on administrative tasks, freeing them to focus on deepening client relationships. In practice, however, advisors experimenting with agentic AI say so far it’s best with narrowly defined use cases — and with humans firmly in control of anything that could introduce compliance or fiduciary risk.
READ MORE: This is the biggest cybersecurity threat for wealth firms
Why data discipline and governance trump autonomy
What makes agentic AI potentially valuable is its ability to orchestrate real business tasks like accessing APIs and interacting across systems, said Meng Tong, CEO of AI software development firm Sotatek, which builds agentic AI systems for finance and other sectors.
Tong said that with the right implementation and governance, his firm has seen agentic AI deliver strong results in structured tasks including pre-meeting client research, portfolio data extraction and analysis, document classification and compliance tagging.
“These are areas where agents can act with high accuracy and measurable return on investment — and where human advisors still retain oversight and final decision-making authority,” he said. “That said, we do not recommend using agentic AI in isolation for discretionary investment decisions or complex suitability assessments — not yet.”
The accuracy of agentic AI depends heavily on data discipline, said Dan Bjerke, president of digital wealth at fintech InvestCloud in New York City. Because AI magnifies poor data just as readily as accurate data, firms must first establish a single, permissioned source of truth before deploying intelligent agentic layers, he said.
“Firms without strong data governance may face correction and rework, reinforcing the need for standardized data models, policy-based access controls and tightly integrated systems before embedding agentic AI into workflows,” he said.
READ MORE: How financial advisors can buy a wealth book of business
How agentic AI is being used on the ground
Advisors who have begun successfully experimenting with this technology say they are doing so slowly, and with limits.
Steven Crane, founder of Financial Legacy Builders in Dayton, Ohio, said he uses forms of agentic AI in his practice, “but intentionally and cautiously.” He said he is not letting it run the business, make decisions or interact with clients autonomously.
“Where it helps is behind the scenes: organizing information, stress-testing assumptions, flagging inconsistencies and helping me think through scenarios faster than I could on my own,” he said.
The biggest benefit so far has been time savings, said Crane.
“Not because AI replaces my judgment, but because it reduces friction,” he said. “It helps me catch things I might miss when I am moving fast and frees up more of my energy for client conversations, coaching and strategy.”
Christopher Hensley, president and CEO of Houston First Financial Group, said he has begun using agentic AI with strict guardrails.
Hensley is currently working with an agentic AI platform that includes a personally identifiable information and compliance firewall, which was “non-negotiable” for him as a fiduciary. He has limited his use to read-only tasks that analyze and surface insights from his data.
“It does not take autonomous actions that affect client accounts or client-facing communications,” he said.
Keeping the human in the loop
Even with these gains, Crane said he feels agentic AI is not yet reliable enough to be trusted without human oversight. He said he still reviews, corrects and sometimes completely disregards what it produces.
“If someone blindly plugs AI outputs into their workflows, they are asking for problems,” he said. “In some cases, fixing bad AI output can take longer than doing the work manually.”
Beyond the platform itself, Hensley said he has built a governance layer to make the boundaries explicit. He said he maintains a controller document that defines what the AI can and cannot do, along with a context index that captures how and why he makes certain decisions — especially edge cases and exceptions — so the system reflects real judgment rather than generic best practices.
“In practical terms, the AI proposes and I approve,” he said. “Trust is earned over time, not assumed.”
Agentic AI is reliable today when it’s used with a human in the loop and constrained to the right jobs, said Hensley. He says he does review and occasionally corrects outputs, but he feels that the review process is far faster than starting from scratch.
“In practice, it feels less like rework and more like editing a capable junior associate’s work,” he said. “Where I would not rely on it without oversight is anything compliance-sensitive, nuanced client guidance or any workflow that could introduce unintended client-facing language.”
Hensley said his advice to other advisors is to start with low-risk tasks, build a clear governance framework and resist the urge to over-automate too quickly.
“Used this way, agentic AI can already function as a force multiplier without introducing unnecessary compliance risk — and experimenting outside the client workflow is often the safest way to learn what’s viable before bringing anything ‘inside the walls,'” he said.
In the future, Crane said he sees agentic AI becoming a powerful tool, but not a replacement for advisors.
“It works best as a second set of eyes, not a decision-maker,” he said. “Advisors who treat it as an assistant will win. Advisors who treat it as a substitute for thinking will eventually lose client trust.”
