In brief
- Most AI tools available to firms today are sophisticated question-answerers: you ask, they respond, and the work of applying that answer is still yours.
- A genuine AI agent is different. It does the work directly, running a return, working through an audit, reconciling accounts, drafting a memo, while pulling from the right sources and flagging where a person needs to step in.
- The value of an agent isn’t full automation. It’s knowing its own boundaries and handing the work back to a person at the right moment.
- Firms getting the most out of this technology start by asking where their best thinking is bottlenecked, then apply agents there specifically.
- The result isn’t a smaller team. It’s a model where a junior working alongside an agent produces higher-quality work, and a senior’s judgment reaches further instead of being the bottleneck.
The gap between answering and doing
Everyone in this profession is talking about AI agents right now, but the term isn’t clear. Most of what firms are using today, however capable, still falls into the category of a sophisticated question-answerer. You ask a question, it responds, and you’re the one who takes that response and turns it into finished work.
Agents are a different category of tool. Instead of answering a question about a tax return, they work through the return itself. Instead of summarizing what an audit procedure might involve, they work through the audit. They reconcile accounts. They draft the memo. Along the way, they pull from the right sources, flag where a person’s direction is needed, and verify where it isn’t sure. The distinction is less “search engine” and more “coworker that knows when to act and when to ask.”
Agents that know when to ask are the critical piece.. It would be easy to build an agent that simply runs to completion on every task and hands you a finished-looking answer. That isn’t where the real value sits. The value is in an agent that knows what it can’t discern and passes the work back to a person at exactly the point where judgment, not execution, is what’s needed. An agent that can’t tell you when it’s uncertain isn’t saving your team time, it’s shifting risk onto whoever reviews its output.
Getting knowledge transfer right with AI
The smartest people in this profession have spent decades developing judgment that lives almost entirely in their own heads. That kind of discernment has always been difficult to transfer to the rest of a firm. It usually depends on a client getting matched with the right partner or senior at the right time, and clients who don’t get that pairing simply get a different level of service.
For the first time, there’s a real path to making that expertise accessible to more people across a firm, not just the clients fortunate enough to land with the right person. But that only happens if firms are deliberate about where they point the technology. The firms getting this right aren’t starting with a tool and looking for a use case. They’re starting with a question: where is our best thinking bottlenecked? Once they’ve found those places, that’s where agentic tools, the kind that can actually do work rather than only answer questions, start to make a measurable difference.
What this means for how firms staff and develop people
There’s a reasonable worry underneath all of this: if agents can do more of the work, does that mean firms need fewer people, particularly at the junior level?
The evidence from firms actually deploying this well points the other direction. The traditional staffing structure doesn’t shrink, it expands. A junior working alongside an agent produces higher-quality work than they would working alone, because the agent is handling more of the mechanical execution and flagging uncertainty rather than guessing through it. A senior, in turn, stops being the person everything has to pass through to get done. Their judgment becomes the multiplier instead of the bottleneck, reaching more clients, more engagements, and more of the work that actually requires their experience.
That shift has a direct economic effect. Practitioners with more bandwidth to focus on high-impact work can take on more complex advisory scenarios, anticipate what their clients need earlier in an engagement rather than reacting to it, and deliver guidance that’s sharper because it’s coming from someone who had the time to actually think it through. Firms that make this shift well aren’t cutting costs by doing less. They’re growing capacity by doing more of what their best people are actually good at.
The practical takeaway
For firm leaders evaluating this technology, the useful test isn’t whether a tool is labeled “agentic.” It’s whether it can tell the difference between a task it can complete on its own and one where a person’s judgment needs to take over, and whether it’s honest with you about which is which. Tools that can’t make that distinction are compounding liability, creating a system that’s worse than each of its parts.
The firms that will benefit most from this next wave of technology are the ones asking a narrower, more useful question than “how do we adopt AI.” They’re asking where their best thinking is currently stuck, and building from there.
David Yue is co-founder and CEO of Accordance, an AI platform built for tax research, founded by AI researchers who spent their first years working alongside tax leaders, professors, and firm partners. Learn more at accordance.com.
Sign in to get access to this free resource, and all of our whitepapers and reports.
Download this content today!
Register Now Already registered? Click here to Log In
