Accounting firms are letting staff build AI tools. Now comes the messy part.

Accounting Today has a good example of where firm technology is heading next. Armanino moved from a single firmwide innovation group to a more decentralized model where each business unit has its own AI expert watching what people are building inside real client-service workflows.

That sounds like a small org-chart tweak. It is not. It is the difference between treating AI like a demo and treating it like an operating system.

The best ideas are coming from the people buried in the work

The source story describes a pattern every firm owner will recognize. A staff member builds a useful tool for one narrow pain point. A teammate asks to use it. Then a few more people ask. Suddenly 30 people are relying on something that started as one person's side project.

That is great until the tool breaks, starts drifting, creates a data issue, or turns into a support job for the person who built it. The employee had the idea because they understood the work. That does not mean they should own security, cloud spend, maintenance, QA, and firmwide rollout forever.

This is the quiet governance problem inside the AI boom. Firms want staff to experiment because the best workflow fixes usually come from the people closest to the mess. But once a tool touches client data, workpapers, billing, board summaries, tax documents, or firm reporting, it stops being a toy.

Seventeen versions of the same tool is not innovation

Armanino's Carmel Wynkoop told Accounting Today that before the firm formalized its approach, many people were building similar solutions without knowing it. Her example was blunt: the firm did not need 17 people building sales agents.

That is the part many firms will miss. AI sprawl does not always look reckless. It can look productive. Everyone is excited. Everyone is saving time locally. Everyone has a clever workflow. Then leadership realizes the firm has duplicate tools, unknown data paths, unsupported scripts, and recurring cloud costs attached to experiments nobody is using anymore.

Armanino says it built a cloud spend analysis tool and saved $700,000 by turning off unused services. That number is the story. AI adoption will not just be measured by how many tools a firm launches. It will be measured by how many survive governance, reduce real work, and avoid becoming expensive clutter.

The CPA firm version of AI governance is practical

This does not need to become a 90-page policy nobody reads. The basic questions are simple. Who built the tool? What client or firm data does it touch? Who reviews the output? Who owns maintenance? What happens if it fails during busy season? Is there a cheaper or safer version already inside the firm?

Firms that answer those questions early can let staff keep experimenting without turning the place into a pile of unsupported internal apps. Firms that avoid the questions will eventually learn the hard way that AI pilots have carrying costs.

The better move is not to shut down staff-built tools. It is to create a path for the good ones to graduate. Business-unit experts can spot the workflows worth scaling, central teams can enforce security and cost controls, and firm leaders can decide which experiments deserve real ownership.

What CPAs should watch

This matters beyond Armanino. Every firm that tells staff to use AI is also creating a shadow software shop. That can be a strength if the firm has guardrails. It can become a liability if nobody knows what is being built.

The next competitive edge may not be which firm buys the flashiest AI platform. It may be which firm turns staff ideas into governed workflows without killing the creativity that made those ideas useful in the first place.

Source: Accounting Today