AI didn't take the tax planning work

I've had roughly the same conversation on repeat for the past couple of years, with partners, tax managers, and candidates across California. AI is suddenly on everyone's mind, and underneath the question there's always the same one: what does this mean for my job?

The adoption numbers moved fast, even by tech standards. The Thomson Reuters Institute's 2026 AI in Professional Services Report, which surveyed more than 1,500 professionals, found organization-wide generative AI use nearly doubled year over year, climbing from 22% to 40%. Only 19% of organizations say they have no plans to adopt. Among tax professionals who do use it, 86% are using it at least weekly and 36% reach for it multiple times a day. Tax research is now the single most common use case in the profession, with document review and return preparation close behind.

So it's real, and it's fast. It just isn't doing what people feared it would do.

What it's actually taking

A candidate told me last year she was worried AI would take over her tax planning work. Eighteen months on, what's actually happened is more boring and more useful.

AI is chewing through document extraction, first-pass data entry, transaction categorization, anomaly flagging, and drafting workpapers. The unglamorous, high volume, low judgment end of the work. Thomson Reuters' own framing of where 2026 is heading is a shift from compliance-focused delivery toward advisory, driven by exactly this: automation eating preparation time first.

The size of the prize is smaller than the vendor marketing suggests, though. In Thomson Reuters' Future of Professionals research, practitioners estimated AI would save them an average of around five hours a week. That's real money over a year, and it's nothing like the transformation some firms are being sold.

The part nobody mentions: you may not be able to keep the savings

Here's the finding I didn't expect, and it complicates the standard "free up hours, sell advisory" story.

On June 24 this year, the IRS Office of Professional Responsibility issued Alert 2026-19, its first formal guidance on generative AI in federal tax practice. It doesn't create new rules. It maps six existing Circular 230 obligations onto AI-assisted work.

One of those is Section 10.27(a), on fees. The OPR's position is that billing for manual time you didn't actually spend, or double-billing work the AI did, can amount to an unconscionable fee depending on the pattern and the scale of the gap. Cost savings from AI are expected to reach the client.

Read that alongside the advisory pivot everyone is planning and you get a more honest picture. The hours AI gives back don't automatically convert into margin on the same compliance work. They convert into capacity, and capacity is only worth something if the firm has higher-value work to put into it. Firms that have built an advisory offer will do well here. Firms that haven't will find they've bought efficiency they can't bill for.

AI doesn't sign the return

Every serious conversation I have with a managing partner comes back to the same place, and the OPR guidance now says it in plain language: technology is a powerful tool, not a substitute for professional judgment.

Under Section 10.22, practitioners have to review AI-generated content properly before it goes to a client or to the IRS, verifying facts, citations, and calculations. The OPR points to courts sanctioning lawyers over fabricated AI citations and warns tax practitioners they carry the same exposure. The AICPA's position runs parallel: competence, due care, and confidentiality don't dissolve because new technology is involved, and that responsibility can't be delegated to a machine.

A large language model can draft an 1120S. It can't take professional responsibility for it, defend the position under examination, or read a room in a client meeting when the client's business is falling apart. That accountability sits with a person, and now it sits there with a regulator's guidance pointing directly at it.

What this means if you're in a compliance-heavy seat

If your current role is close to pure preparation, with minimal client contact and minimal review responsibility, that's the part of the job getting automated fastest. The work still needs a human final check, and the OPR has just made that check a formal obligation rather than an assumption. But the volume of hands-on-keyboard time on routine returns is shrinking, whether your firm has said that to you directly or not.

The signing, reviewing, advising end of the work isn't just safer. It's the end the regulator has now explicitly anchored to a human. That's worth factoring into where you aim your next move.

Not that the job disappears. The center of gravity in it has moved, and the people who move with it early will have an easier few years than the people who wait to be told.

AI is only half the story, though. The other half is happening on the staffing side, in offices most California firms have never visited, in Bangalore and Manila. That's the next piece.

Sources: Thomson Reuters Institute, 2026 AI in Professional Services Report and Future of Professionals Report; IRS Office of Professional Responsibility Alert 2026-19, Introductory Guidelines for Responsible AI Use in Federal Tax Practice, June 24, 2026; AICPA Code of Professional Conduct guidance on AI and ethics.