The Two Companies That Know Models Best Just Went Into Consulting

When I started my business in 2024, I told people I was going to do coaching. I'd like to report that I had a considered reason for it. I didn't, and the truth is worse: I did not actually know what coaching was.

It took me about a year to work out that I'm not a coach at all. Coaching draws the answer out of the person across from you, and the people who are good at it have a patience I don't have. I'm not especially skilled at getting someone to discover the truth within themselves. What people hire me for is the opposite motion: I go read everything, chase down the numbers, figure out what's actually true, and then tell them what I found and what I'd do about it. They want the expert in the room, and they want the work behind the expertise. Once I owned that about myself and about my value, it felt like the lights turned on and the business started to flow.

I still catch myself asking a client a leading question and waiting for them to arrive at a conclusion I already have written down in a document three tabs over. Bad habit from a job I never actually had.

The thing that took me a year to learn is the difference between selling an insight and selling the work of getting to one. Two companies with far better information than mine appear to have spent this year learning the same lesson, at a scale that makes my version look like a rounding error.

The two new AI consulting firms, and what they actually raised

In May 2026, both of the leading model developers moved to stand up services businesses with outside money. Reuters reported that OpenAI was raising roughly $4 billion from 19 investors, including TPG, Bain Capital and Brookfield Asset Management, for a joint venture called The Deployment Company. Anthropic made a similar move, raising $1.5 billion from investors including Blackstone, Hellman & Friedman and Goldman Sachs. Reuters attributes that second figure to Wall Street Journal reporting rather than its own, so treat it as one step removed. Both ventures, per Reuters' sources, were then in talks to buy AI services firms outright.

One note on that first figure, because I nearly got it wrong myself. A vendor blog post I came across put OpenAI's vehicle at more than twice the committed capital Reuters reports. Roughly $4 billion is the number I'm using, because between a wire service citing named investors and a marketing blog with something to sell, I'll take the wire service. The inflated version was off in the direction that makes the story sound bigger, which is usually the direction these errors run.

The Anthropic firm got a name in July. TechCrunch reported that it launched as Ode with Anthropic, a $1.5 billion AI implementation company, and that it competes not only with The Deployment Company but with Deloitte and Accenture, both of which have built their own forward-deployed-engineer teams. The structure is the part that tells you how much lab expertise had to travel with the money. Blackstone's announcement and Anthropic's own newsroom post both describe a standalone entity with Anthropic engineering and partnership resources embedded directly within its team, with applied AI engineers from Anthropic working alongside the firm's own. Those two are the participants describing their own deal, so read them for structure and not for market analysis.

The labs went into services because deployment is where their revenue gets stuck

Anthropic's CFO, Krishna Rao, put it as enterprise demand for Claude "significantly outpacing any single delivery model," with the new firm bringing "additional operating capability to the ecosystem and capital from leading alternative asset managers." Blackstone's Jon Gray said hiring highly skilled workers would "break down one of the most significant bottlenecks to enterprise AI adoption." Both sentences describe the same shape of problem, which is that the product works and the customer still can't get it running.

Reuters names the mechanism plainly: companies "need engineers and consultants to tailor AI models to their specific data, systems and workflows, and to adapt the software as business needs change." That second clause is the part that turns a project into a practice. Data changes. Systems get replaced. A workflow that made sense in March is wrong by September, and somebody has to notice and adjust. That's not a deployment, it's a standing relationship, and standing relationships are the thing consultancies have always sold.

Where the value in AI work actually sits

The sentence I keep coming back to is Reuters' own, not mine: "what is often cast as a high-margin software business that could eliminate the need for consultants still depends on labor-intensive, highly skilled services."

Neither lab has said that models can't replace consultants, and I'd expect both to argue the opposite in a keynote. What they've done is allocate billions of dollars, and capital allocation is a more honest statement of belief than a keynote is. My reading is this: if the two organizations on earth with the deepest possible knowledge of what these models can do have concluded that the durable margin sits in labor-intensive human implementation, then the confident claim that AI is about to replace consultants owes the rest of us an explanation. The people with the most information and the strongest incentive to believe otherwise just voted with their balance sheets.

None of this contradicts what the numbers already showed. It fits with what the market repriced when it punished the incumbents. The market didn't decide that expertise was worthless. It decided that a particular way of packaging and selling expertise had gotten expensive relative to what it delivered, which is a different and much more survivable finding.

How this reshapes the market between the Big Four, boutiques and independents

The competitive pressure on a small practice just changed shape, and it did not get simpler. If you run a boutique advisory firm or a solo practice, the threat in this story isn't that a model will do your thinking. It's that two implementation firms now exist with private-equity money behind them and a warm path to thousands of companies.

Goldman Sachs' Marc Nachmann described the venture as enabling mid-market companies to deploy Anthropic's AI solutions, and said that "by democratizing access to forward-deployed engineers, the new company can help the expansive network of portfolio companies in our Asset Management business." Anthropic's own announcement says the firm will work with mid-sized companies across sectors, initially through those investor portfolio companies. So the entry point is mid-market, and the introduction comes from the board.

That's distribution, not capability. It's the same advantage the Big Four have always had, rebuilt from scratch in a year with cleaner technology and a better story, and pointed at exactly the segment that used to be too small for a Big Four engagement and too big to ignore. This sits right alongside the market splitting and next to where the real threat was all along, which was never the model and was usually somebody with an existing relationship.

A different threat calls for a different response, and the response isn't to get faster at prompting. It's the thing you're already better at than a firm parachuting in through a portfolio introduction: knowing the business, the sector, and the specific person whose job gets harder if this goes wrong. Nobody at scale is going to out-relationship you. They will out-distribute you, so the work is to be findable and specific enough that the introduction comes to you too.

What to ask any AI consulting firm you're considering

Ask who does the work after the launch, and get a name. The Reuters mechanism tells you the value shows up in adaptation over time, so a proposal that ends at go-live is priced for the easy half. Ask what happens when the underlying data or the workflow changes, and listen for whether the answer involves a person or a change order.

Ask what they will tell you not to pursue. A firm whose economics depend on billing implementation hours has a structural reason to find implementable work, and the useful advisors are the ones who will spend a call talking you out of something. That's the Cheap Yes in its natural habitat: fast, agreeable, credentialed, and expensive later when it turns out to have been wrong. It's also worth asking who checks their outputs, because what the incumbents got wrong in public this year was verification, not ambition.

And ask where the model came from. Not because vendor alignment is disqualifying, but because it tells you which recommendations are structurally available to them. My own tooling runs on Claude, which is a preference I'll defend on the merits and also a thing you should know about me before you weigh my read on Anthropic's business.

The version of this I lived was much smaller and took much longer to admit. I sold coaching for a year while people quietly kept hiring me for research and judgment, and the business only worked once I described it the way the buyers already understood it. When I finally said it out loud on a call with other consultants, one of them told me I was the first person who'd ever drawn that line for her, which suggests I wasn't the only one confused about what I was selling. The labs are making the same correction with a great deal more capital and a great deal less delay. Everybody's insight got cheaper this year, mine included, and the work of turning an insight into something a business can actually run on did not get cheaper at all.

If you're rethinking how your own practice is packaged against all of this, email matthew@fieldway.org and we can talk it through.

Sources

• Reuters, OpenAI, Anthropic ventures in talks to buy AI services firms, sources say (5 May 2026)
• Anthropic, Building a new enterprise AI services company with Blackstone, Hellman & Friedman, and Goldman Sachs
• Blackstone press release
• TechCrunch, Anthropic, Blackstone bet the next trillion-dollar AI business is implementation, not models (15 July 2026)