Who Trains the Judgment When AI Does the Work First?

This was the hot topic on our last AINA Advisory Council call: if an AI agent is doing the work a first-year associate used to do, and the new grad’s job becomes reviewing and managing what the agent produced, where does that new grad ever get the context to know whether the agent got it right? Everyone on the call agreed this is a real problem and they are seeing it profession-wide.

I’ve personally been circling this idea for awhile actually, in my articles on Innate Intelligence & Capability Atrophy, fast-forward and the advisory council conversation reaffirmed my concerns. Capability atrophy is what happens when a skill weakens from disuse, the same way a muscle does when you stop using it, and the tricky part is that you don’t notice what you’ve lost until the moment you suddenly need it. The chess analogy: when a player is handed the engine’s best move instead of working through the position themselves, they skip the productive struggle that actually builds the pattern recognition a strong player relies on later. That’s what’s happening to a lot of junior staff right now, except the stakes are a client’s tax return or an audit opinion instead of a chess game.

AI is very good at producing the “what,” and increasingly capable of explaining how it got there. But an explanation is not the same as judgment. Judgment requires context, accountability, and the ability to decide which risks and competing priorities actually matter. Humans build that capacity through reasoning that’s creative and inferential, not just computational, and that capacity has always been developed the way most expertise gets built: by doing the work first, getting it wrong, and figuring out why. If the agent is doing that work now, the new associate isn’t skipping the boring part of the job. They’re skipping the part of the job that used to teach them how to think like a professional. They can produce a polished-looking deliverable and still not be able to defend the reasoning behind it, because they were never in the room where that reasoning got built.

So the real question for firms isn’t whether to let agents handle first-draft work. That ship has sailed, and there are good reasons to let it. The question is what replaces the productive struggle that used to happen by accident, when a first-year spent three hours getting a schedule wrong before a manager showed them why. If we don’t design something deliberate to take its place, we’re going to end up with ladderless organizations, firms where the rungs junior people used to climb on their way to expertise have quietly been removed.

Here are four design principles that firms can begin testing now:

1. Simulated reps: Create realistic scenarios where emerging professionals must make the decision before seeing the AI recommendation.

2. Teach-back reviews: Require reviewers to explain and defend the output, not merely approve it.

3. Graduated autonomy: Let junior associates complete selected work manually or with limited AI assistance before moving into agent supervision.

4. Exposure to consequences: Bring junior staff into client conversations, review meetings, and exception handling so they see why decisions matter.

Advisory Council Insights:

• Arianna Campbell: “The first shift has to be mindset. Firms have to stop assuming emerging talent can only build judgment the same way previous generations did. Traditional early-career work mattered because of what people learned through it: how to recognize patterns, think critically, and understand the impact of their decisions. If AI changes the work, firms have to be much more intentional about creating those learning opportunities in new ways. The goal is to preserve the learning, not the tasks.”

• Randy Johnston: “Most AI products for the profession have developed a form of “Transparent AI” that explains how they achieved the result and shows the work steps. This methodology has promise for educating professionals in Tax, Audit, and CAS, and potentially other areas of practice, as I discussed with a software provider yesterday. This may be the most promising development on upskilling new professionals.”

• Tom Hood shared one place this is already moving. Through its Profession Ready Initiative (PRI), the AICPA is developing simulations to help new professionals build the judgment and context they’ll need to work effectively in an AI-native environment—including overseeing AI agents performing work they once would have learned by doing themselves.

If agents can show their sources, surface assumptions, flag uncertainty, and explain the steps behind an output, then review can become part of the training ground. But transparency alone will not solve this. Firms will also need to create the learning experiences, exposure to client conversations, and protected opportunities for emerging professionals to wrestle with problems before the answer appears.

Otherwise, we may build firms that can produce more work than ever, and fewer people capable of standing behind it.

— Kacee Johnson, Executive Director

The 2026 AI Native Accounting Award Winner for Small Firm AI Strategy & Governance is highlighted in our first case study.

If you want to see what deliberate, governed AI adoption looks like at a genuinely small firm, we just published a case study on Financial Optics, a three-person restaurant accounting and advisory practice managing 18 clients at the controllership level. A few of the pieces worth pulling out:

· A formal AI acceptable use policy, covering three tiers of data sensitivity with mandatory human review before anything reaches a client

· 90-day, role-specific onboarding plan so both team members ramped up on AI deliberately instead of ad hoc

· Claude-powered workflow that turns raw QuickBooks and Fathom data into client-ready executive summaries with month-over-month analysis

· No added headcount, even as the firm expanded what it delivers to all 18 clients

Owner Tim Sernett: “I stopped looking at AI as a means of saving time, and started looking at it as a means of expanding capabilities.”

Click here to access the full case study >>>

We’re awarding two $5,000 scholarships this year to students who can show real, working AI use cases, not just essays. Open to undergrad and master’s students in accounting, corporate finance, or taxation at accredited U.S. institutions. Winners will have a podcast feature, six months of mentorship with industry leaders in addition to the $5k award. Applications close October 31, 2026. If you know a student who’s thinking about how AI will impact future practitioner skillsets, send them to the scholarship page at ainativeaccounting.org.

Thank you to our Founding, Strategic and Ecosystem Partners for making these scholarships possible. Their support is funding the investment into future practitioners.