Last week, Anthropic announced a $1.5 billion joint venture with Blackstone, Goldman Sachs, and Hellman & Friedman. The stated goal: embed Claude and Anthropic engineers directly inside mid-market companies. Not sell them software. Not run a pilot. Actually embed — forward-deployed engineers sitting inside the client.
OpenAI is doing the same thing with TPG and Bain. Sequoia is calling the product “autopilots” — AI that sells the work, not the tool. And the market they’re targeting? A $300 to $400 billion management consulting industry.
The press is running with the obvious story: AI is coming for consultants. But a professor at Cardiff Business School named Joe O’Mahoney — who studies consulting M&A and equity value for a living — just published a piece arguing the obvious read is completely wrong. His actual argument is more interesting. And I think it’s more important for anyone in the accounting and finance world.
Why Consulting Was Always Hard to Crack
To understand O’Mahoney’s thesis, you need to start with a simple question: why would anyone want to enter the consulting market?
By software standards, it’s a terrible business. Fragmented. Relationship-bound. Low-margin at scale. Notoriously resistant to standardization. And in Western markets, it’s basically flatlined over the last few years.
Previous technology waves have tried to crack it and mostly bounced off. RPA tools automated workflows. Analytics platforms gave firms better dashboards. AI writing tools sped up report production. But none of them fundamentally changed who was doing the high-value advisory work — because at the top of the market, people buy high-stakes advice from people they trust. They buy it at 3am when something goes wrong. They buy it because a partner’s name is on it.
Think about it this way. You can automate a medical imaging scan. You can use AI to flag anomalies. But when a patient is sitting across from a doctor getting a serious diagnosis, they’re not asking for the most efficient algorithm. They want a human being who can read the room, weigh the specifics of their situation, and stand behind the recommendation with their license and reputation. That’s legitimized judgment — and it’s exactly what technology has struggled to replicate.
What Anthropic Is Actually Buying
So why is Anthropic writing a $1.5 billion check into a terrible market?
O’Mahoney’s answer: they’re not buying consulting. They’re buying proximity to enterprise data and enterprise decision-making — the two things foundation models cannot yet train on, and the two things that determine who wins the next phase of AI.
Read the JV structure carefully. Engineers embedded inside the client. Operating systems of record. The press release calls it a services firm. O’Mahoney says it isn’t. It’s a data and decision-rights acquisition vehicle dressed as a services firm.
Think about what consulting firms actually do inside a client engagement. They’re not just producing slides. They’re in budget meetings. They’re in the room when a CFO decides how to restructure the balance sheet. They have access to messy, contextual, behind-the-firewall data that never shows up in a press release or a 10-K. That’s exactly what the frontier AI labs are running out of. Public internet data is exhausted. The remaining alpha — the data that will train the next generation of models — is locked inside enterprises.
McKinsey has spent 90 years getting legitimized access to that data and never figured out how to monetize it beyond the engagement. Anthropic just bought a shortcut, with Goldman’s relationships and Blackstone’s portfolio as the on-ramp.
And even if that data stays behind firewalls and never directly trains the models — O’Mahoney acknowledges that’s possible — it doesn’t really matter. Because the end goal isn’t the data layer. It’s the judgment layer. The tech firms want to own the advice itself. Not just the analysis. The actual recommendation a board acts on.
Who’s Actually Most Exposed
Here’s the counterintuitive part of O’Mahoney’s argument: the biggest firms are more trapped than boutiques.
The conventional wisdom is that scale wins in AI. So McKinsey and the Big Four, with their global reach and proprietary IP, should be best positioned. O’Mahoney thinks the opposite is true — for two reasons.
First, their cost base is the liability. A big engagement at McKinsey is priced to support a global pyramid — associates, engagement managers, partners, knowledge centers. When an AI-native JV can do the same work for 90% less, McKinsey can’t follow the price down without collapsing its partnership economics. The model forbids it. That’s the innovator’s dilemma, except the incumbent can’t even cannibalize itself.
Second, they can’t credibly go AI-native. The market won’t believe McKinsey is AI-native no matter how much it spends on rebranding. Anthropic owns “AI-native consulting” the way Tesla owned “electric car” by 2018: by being the thing, not performing the thing. Boutiques, by contrast, can make the pivot. A 50-person firm can be genuinely AI-native in 12 months. A 50,000-person firm cannot.
But the firms most at risk, according to O’Mahoney, are the ones in the middle: mid-tier generalist advisory firms with broad service portfolios, days-based billing, and junior-leveraged delivery. His call is direct — most will be undercut by an order of magnitude within 36 months and will be acquired, merged, or quietly wound down by 2030.
What This Means for Accountants
Two practical implications worth sitting with.
The first is diagnostic. What proportion of your billable work is intelligence production — research, analysis, memo writing, structured deliverables? Audit support documentation. FP&A commentary. Deal support memos. Tax research summaries. That’s the work that gets undercut first. It’s the most legible to AI, the most reproducible, and it will face price compression faster than most people expect. If your value proposition is “I produce high-quality analysis efficiently,” that’s the lane these JVs are entering.
The second is about where the defensible position actually is. O’Mahoney calls it the trust layer: legitimized judgment backed by a named individual whose professional credential allows them to stand behind a recommendation. That’s closer to what auditors do than what consultants do.
A CPA license is a structural moat — not because of what you know (AI will close that gap), but because of what you can stake. Your license. Your name on the engagement letter. Your accountability if something goes wrong. There’s a well-documented behavioral pattern here: given a choice between a CPA and an AI firm, most clients will choose the CPA for high-stakes decisions — not because the output is necessarily better, but because the CPA absorbs the risk. Goldman’s relationships and Blackstone’s portfolio don’t come with a CPA number.
Key Takeaways
- The Anthropic-Blackstone JV is not a consulting play — it’s a data and decision-rights acquisition vehicle. Consulting is the wedge; the operating layer of the enterprise is the prize.
- Intelligence-layer work — research, analysis, structured deliverables — faces extreme price compression within 36 months.
- The Big Four and MBB are more structurally trapped than boutiques. Their cost pyramids can’t follow pricing down, and they can’t credibly claim to be AI-native.
- The trust layer is the CPA’s defensible position: legitimized judgment backed by a professional credential that absorbs client risk.
- Ask yourself: what percentage of your billable work is intelligence production vs. relationship-backed judgment? That ratio is your exposure score.
Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: [lesson link]
Leave a Reply