What's new:

In a blind Stanford Law study, 16 U.S. law professors evaluated nearly 3,000 anonymized answers to student contracts questions and preferred AI-generated responses in 75% of head-to-head matchups against answers written by their own peers. Rather surprisingly, they flagged AI answers as pedagogically harmful only 3.5% of the time, versus 12% for the human-written ones. Blind expert reviewers rated the machines both better and safer than their colleagues.

How it works:

Sixteen professors each wrote 40 representative questions students ask after class, drafted their own answers, then graded a blind mix of AI and peer responses without knowing the source. The researchers calibrated AI output to match human answer length and structure, used multiple evaluation methods, and explicitly asked whether each answer might mislead a student. Law was purposely chosen as a subject area, because it rewards synthesizing ambiguity into a defensible position rather than right-or-wrong factual recall.

Behind the news:

Most AI benchmarks live in domains with a correct answer: math, coding, multiple choice. This study deliberately went after the opposite — the latent professional standard practitioners use to judge each other’s reasoning when two opposing arguments can both be right. That’s the exact terrain professionals have claimed as the human moat. The finding doesn’t say AI knows more law. It says expert evaluators, judging blind, found AI’s judgment-laden reasoning better than their colleagues’.

Why it matters:

The accounting profession has leaned on the same defense law just watched buckle: The work that survives automation is the judgment work — the gray-area tax position, the revenue recognition call under ASC 606, the audit estimate that requires weighing competing inferences, the advisory conversation where there’s no single answer. If blind expert reviews prefer AI in a sister profession built on exactly that kind of reasoning, the “clients pay us for judgment, not output” line stops being a guarantee and becomes a hypothesis. Note what this study measured and didn’t: answer quality, evaluated by experts. It did not measure accountability, liability, or who signs the opinion. That’s where your defensible value now concentrates.

Our thinking:

The moat was never judgment in the abstract. It’s judgment plus the professional liability that makes someone answerable for it. AI just demonstrated it can produce reasoning that experts rate higher than their peers’ in a judgment-rich field, which means the firms still selling “trust us, we think harder” are selling something a blind test no longer backs up. The defensible position shifts to what AI structurally cannot hold: the signature, the fiduciary duty, the regulatory accountability, the relationship where a client needs a named human on the hook. Firms that rebuild their value proposition around accountability and client trust — and use AI to do the reasoning underneath — will widen their margins. Firms still charging premium rates for the reasoning itself are pricing a commodity. On its surface, the Stanford study is about law students. In practice, it's about your billing model in 18 months.

What's new:

The Wall Street Journal reports that accounting schools are rapidly redesigning their curricula as AI reshapes the profession. Programs that once updated coursework every three to five years are now making changes every semester, replacing static textbooks with dynamic digital content, incorporating AI simulations, and preparing students for a future career where routine work is increasingly handled by AI. Some Big Four leaders estimate AI could perform 20%–30% of a typical financial audit by 2029, fundamentally changing what new accountants need to learn.

How it works:

For generations, the profession has relied on an apprenticeship model: Junior associates mastered repetitive work before earning opportunities to exercise judgment. AI compresses that progression by taking on much of the routine execution. Universities are responding by teaching students how to evaluate AI-generated outputs, think critically and develop professional judgment earlier in their careers. Schools are also investing in AI-powered simulations that allow students to practice client interactions, inventory observations, and audit scenarios in ways that better reflect the work they'll encounter after graduation.

Behind the news:

This isn't simply a curriculum update, it's a rethink of how students learn to become accountants. The profession has long assumed that judgment is earned through years of performing the underlying work. But if AI performs much of that work, firms and universities must find new ways to develop professional skepticism, critical thinking and decision-making. The conversation is no longer about teaching students how to use AI. it's about redesigning the learning journey so future professionals can build expertise in an environment where AI is doing much of the execution.

Why it matters:

As AI removes more entry-level work, firms can't assume future managers and partners will naturally develop the experience previous generations accumulated through repetition. The firms that intentionally redesign onboarding, mentorship, and professional development around AI will build stronger future leaders. Those that simply automate junior work without replacing the learning opportunities risk creating a generation of accountants with fewer chances to develop the judgment that clients ultimately pay for.

Our thinking:

This article reinforces one of the biggest questions facing the profession over the next decade: If AI performs the work that used to teach accountants how to think, where does judgment come from? Firms often focus on the productivity gains AI delivers today, but the more important challenge is ensuring tomorrow's professionals still develop the expertise that clients can trust. The firms that solve the apprenticeship problem won't just have a recruiting advantage, they'll shape what it means to become a CPA in the AI era.

What's new:

Thomson Reuters released its 2026 Future of Professionals Report, surveying more than 2,200 professionals across accounting, legal, tax, corporate risk and government. The headline isn't that AI adoption is accelerating, we already knew that — it's that execution isn't. While 74% of professionals now use AI several times a week, more than a third say their organization's AI strategy hasn't translated into how work actually gets done. The result is a widening gap between executive ambition and frontline reality.

How it works:

The report separates AI adoption from AI transformation. Firms are writing AI strategies, buying tools and communicating vision, but many haven't redesigned workflows, trained their people or embedded AI into day-to-day operations. That disconnect creates predictable behavior: Professionals find their own tools. Thirty-four percent of respondents admit they're using unsanctioned AI applications, while 41% still lack access to AI tools designed specifically for professional work.

Behind the news:

Shadow IT and AI use is the next phase of AI adoption, and probably one of the biggest security risks to firms. Twelve months ago, firms worried whether employees would use AI. Today, employees are already using it, oftentimes without permission. The bottleneck has shifted from technology to organizational change. The firms creating durable advantage won't necessarily be the ones with the best models, they'll be the ones that redesign how workflows through the firm, establish governance that people actually follow, and give professionals tools they trust enough to stop reaching for consumer AI.

Why it matters:

Accounting firms often assume their AI strategy exists because leadership has approved one. Employees judge it differently. If the tools aren't available, the workflows haven't changed, or the training never happened, there is no strategy from their perspective. Meanwhile, clients are raising expectations. Nearly 80% say AI-enabled quality improvements are important when selecting professional service providers, yet only 6% believe most firms are delivering on that promise. AI is quickly becoming a competitive differentiator, not because clients care what model you use, but because they expect faster, better and more proactive service.

Our thinking:

One thing we've consistently observed across the profession is that AI transformation rarely stalls because the technology isn't ready. It stalls because organizations try to layer AI onto yesterday's operating model. Buying licenses is easy, but reimagining workflows, redefining roles and building new management habits is much more complex. The firms that close the gap between strategy and execution won't simply have higher AI adoption — they'll build organizations where AI becomes part of how work gets done rather than another initiative employees quietly work around.

What's new:

Metaculus, one of the world’s leading forecasting platforms known for aggregating predictions from expert forecasters and tracking their long-term accuracy, found that an AI system outperformed every human competitor in its finance-only Market Pulse tournament — including the forecaster tied for first place in Metaculus’s flagship human-vs-machine contest. One founder at a recent prediction market conference described turning $35 into $2 million on Kalshi in seven months, highlighting how quickly AI-assisted forecasting is moving from academic exercise to real-world financial decision-making.

How it works:

FutureSearch and Preseen build on frontier models — the same GPT and Claude models anyone can use — by adding a scaffold of AI subagents that gather primary sources, evaluate each underlying assumption and only then commit to a final number. Tested live, FutureSearch read sixteen sources in two minutes and delivered a cited, 212-source estimate in five minutes – for $8.

Behind the news:

AI researchers Sayash Kapoor and Arvind Narayanan named forecasting, alongside persuasion, as a task near its human performance ceiling — one AI wouldn’t meaningfully clear. Metaculus’s own tracking shows scaffolded systems closing the gap at roughly 0.9 rating points a month, with parity projected in six months. Finance is already ahead, not approaching it.

Why it matters:

Going concern assessments, impairment testing and valuation work are conjunctive-probability problems (several dependent assumptions rolled into one number) — exactly what these systems solve. A CPA can get a cited, five-minute probability estimate on litigation exposure or refinancing risk for the price of a client lunch. With an AI beating the strongest human forecaster in Metaculus’s finance tournament, the assumption that judgment wins by default is gone.

Our thinking:

Firms that cite an AI probability estimate in every major judgment call — going concern, fair value, litigation reserves — get a documented basis built in minutes, not days. Firms that treat those calls as pure partner judgment, by contrast, will look arbitrary once a regulator asks why the number wasn’t checked against an $8 model. The billable-hours model for scenario advisory is the first casualty, and it sits inside the profession’s own turf — which is why the move is routing these estimates into the workpapers before someone else’s version becomes the benchmark.