Your AI Advisor Is Selling You Their Stack

Three weeks ago, OpenAI launched a consulting firm. The Deployment Company — a majority-owned joint venture that raised over $4 billion from 19 investors, with Bain, Capgemini, and McKinsey signed on as founding consulting partners.

Read that sentence again. The company that sells the models now owns the firm that advises you on which models to buy — and three of the biggest names in consulting are its partners.

Nobody involved is behaving badly. They're behaving structurally. But if you're a CIO evaluating your AI options in 2026, you should be very clear about one thing: the number of large advisors with no stack to sell you has fallen to approximately zero.

The referee count just hit zero

The Deployment Company isn't an outlier; it's the pattern. Consider who's now offering to guide your AI strategy:

Model vendors

OpenAI's $4B consulting venture reaches deep into workflow redesign — it even acquired an applied-AI firm with ~150 forward-deployed engineers to seed it. Anthropic has announced its own enterprise AI services company, backed by Blackstone, Hellman & Friedman, and Goldman Sachs, plus a $100 million commitment to its partner network for 2026. Both give excellent advice. Both will conclude the answer runs on their models.

Hyperscalers

Microsoft, Google, and AWS are hiring forward-deployed engineers at extraordinary rates — the role grew roughly 800% and is the signature AI hire of 2026. These engineers often arrive "free" with your cloud commitment, then quietly become your outsourced AI center of excellence, steering architecture decisions for years. Free advice from someone paid on your cloud consumption is the most expensive kind.

The Big 4

The pitch shifted in 2025 from "we'll build your pilots" to "rent our pre-built agents." PwC's Agent OS has ~25,000 agents deployed across client operations. Deloitte sells Zora AI's "intelligent digital workers." KPMG has Workbench; EY pushed 150 agents to 80,000 of its own tax staff. When your strategy consultant owns an agent portfolio, guess what your strategy will feature.

Each of these players is competent and useful — as a supplier. The problem begins when the supplier also holds the pen on your strategy.

Why this matters more than it did last year

Vendor bias in IT advice is ancient. Three things make it more expensive now.

First, the decisions lock in longer. Agentic architecture — how agents authenticate, access data, hand off to humans, and get monitored — embeds into your core workflows. Whoever shapes those choices shapes your switching costs for a decade.

Second, the honest answer is usually multi-vendor. Different models genuinely win at different tasks, and prices move quarterly. A portfolio approach is obviously correct and structurally unavailable from anyone who fields their own platform.

Third, the failure data is in. MIT found 95% of GenAI pilots deliver no P&L return, and Gartner expects over 40% of agentic projects to be canceled by end-2027 — often over cost and unclear value. Those numbers were produced with all this expert help on the field. More conflicted advice is not the missing ingredient.

Four questions to ask anyone advising you on AI

You don't need to fire your vendors — you need to know which hat they're wearing. Ask, in writing:

1. Who pays you, besides me? Referral fees, resale margins, alliance incentives, cloud-consumption kickbacks — all of it.
2. What do you sell that could appear in your recommendation? A platform, an agent library, an accelerator, a hosting relationship.
3. When did you last recommend a competitor's stack over your own? Ask for the example. Watch the pause.
4. Will you put stack-neutrality in the engagement letter? The confident referee signs. The salesperson explains why that's complicated.

None of these questions are rude. Any advisor worth their rate has heard them and has clean answers. The ones who bristle are answering a different question.

Keep the players, hire a referee

The practical model isn't vendor abstinence — you'll buy from these companies, and you should. It's separation of powers: the parties who profit from what you pick don't get to specify what you pick.

That referee function can be an internal architecture group with real authority, an independent advisor, or both. What it cannot be is the vendor's own consulting arm, however brilliant. We referee, they compete — that's the arrangement that keeps everyone honest, including us.

Where to start

Before your next AI vendor meeting, write one page: every AI advisor and supplier you're paying, what they've recommended, and what they earn if you follow the recommendation. The overlaps will jump off the page.

And if you want a stack-neutral baseline of where your program actually stands, Delzey's free AI Readiness Score at /readiness is 20 questions and about ten minutes — scored against a framework, not against anyone's product catalog. Nobody's inventory depends on your answers, which these days is a feature worth stating.