“Better to be an investor and not even take all this execution risk.”

That line is from Satya Nadella. He wrote it on July 13, 2022, in an internal email to his Microsoft leadership team. It sat under lock for nearly four years until the Musk v. Altman trial put it in the public record this week. I read it on Tuesday and could not put it down.

The CEO of Microsoft, with $200B in cash and the most complete operation on the planet, telling his own people he would rather be an investor than take on the execution risk of the OpenAI deal. The exact line he used to describe where he stood: “Right now we are a very thin layer on top of NVIDIA and all the IP is with Open AI.” A thin layer on top of someone else’s stack.

The strange part is when the email surfaced. This same week Thomas Kurian announced Google Cloud is hiring additional Forward Deployed Engineers paired with a $750 million ecosystem commitment to accelerate agentic transformations across its 120,000-partner network. And on X, Jaya Gupta documented the move while Maurizio (@themgmtconsult) named the consultant → FDE shift “the great rebrand of the 2020s.”

The two things are the same thing.

The job Google will never fly to Tegucigalpa

Here is the thesis IQ Source has been arguing for months and that this week became public consensus: the value built on top of the model layer is not delivered by the model. It is delivered by the human embedded at the client, translating model output into real processes. That human, in 2026, is called a Forward Deployed Engineer. In 2015 the card said digital transformation strategist. In 1995 it said SAP consultant at Andersen Consulting.

What IQ Source does is occupy that function — the FDE function — for mid-market companies in Central America. The Big Tech FDE economics do not work for a 200-to-300-person operation in the region: embedding an FDE costs the same in Tegucigalpa as in Manhattan, and the per-account revenue does not compare. It is not that they don’t see you; it is that they will not reach you. The reasoning layer either gets built with a partner who already lives in this stack, or it gets billed two years later by a Big Four practice arriving to “discover” what your team already knew.

There are two chained services. AI Maestro is the two-month discovery phase: the same work an Anthropic FDE does in its first eight weeks inside a Fortune 500 client, sized for a company of 80 to 800 people. Real process mapping (not the manual version), AI Opportunity Score per process, an honest Go/No-Go gate at the end. Sometimes we walk out telling a client that no, agents do not belong in this operation — that is part of the value. Technology Partner is the build phase: integrations into your HubSpot or legacy SAP, evals, hallucination guardrails, change management.

It is the function, not the label, that costs money. And the function is not covered by a Google welcome email.

The job that gets relabeled every fifteen years

I have been in computing since 1990, and I have watched this trade switch labels twice before. In the 90s it was the Andersen Consulting junior who lived two years in Sheraton hotels installing SAP. After Andersen collapsed in 2002 the role reappeared as digital transformation lead at Accenture Digital and Deloitte Digital, now selling cloud migrations and CRM redesigns with a new business card. The 2026 FDE is the third version of the same job. The function never changed: embedded at the client, translating between technology and business, shipping monthly. What changes with each rebrand is the prestige ceiling of the trade. This time it left McKinsey and moved to Anthropic, OpenAI and Google.

Maurizio (@themgmtconsult) captured it in one line: “Consultants → Forward Deployed Engineers — the great rebrand of the 2020s.” Breakfastmaxi (@pranavsf) named it the “mckinsificatoion [sic] of software.” What matters is not the label. The prestige ceiling moved, and that reassigns young talent at a speed two cycles ago could not match. Jaya Gupta framed it as a single question: who makes AI implementation feel like the frontier of the trade?

The part Satya did not want public

The email has more than the thin layer line:

“Overall I want us to own — the silicon, infra, foundational model IP and ‘know how’… we have a P&L that will lose 4 bil next year!!! I have not seen anything like this in my 30 years in our industry… Better to be an investor and not even take all this execution risk!”

In a separate email introduced at the same trial, Nadella compared the OpenAI deal to IBM outsourcing the operating system to a small Redmond software shop in the 1980s. He was afraid of becoming “the next IBM” while OpenAI became “the next Microsoft.” Six months later, another $10 billion went in. The bet worked: today Microsoft holds a $228B stake in OpenAI and runs a $37B annualized AI revenue line. Aakash Gupta summed it up this week: “every vulnerability Satya identified in July 2022 still exists. The bet paid off anyway because the alternative was worse.”

That is the asymmetry the rebrand hides. The CEO of Microsoft, holding the biggest negotiating position on the planet, ended up signing a deal he was complaining about in capital letters in private. When everyone celebrates FDE prestige, no one reminds you that the FDE shows up to install Satya’s thin layer on top of your operation. The diligence he would have wanted in 2022 — knowing exactly what he was buying and where the lever actually lived — is the same diligence any mid-market Central American company skips when it signs with an AI partner in the euphoria of the first demo.

The 87% that a Google FDE does not fix alone

Karthik Subramanian (@KkarthikS), a CTO with real enterprise deployments under his belt, published his field notes this week on the OpenAI Deployment Company (the new $4B subsidiary valued at $14B). His critique in one line: “the incentive structures and messy data reality behind the hype.” He cites the MIT study showing an 87% failure rate in enterprise AI projects — a number that does not budge because the FDE wears a Google T-shirt instead of an Accenture one. It budges when someone maps real processes before the first agent, identifies where the data is dirty, and tells the client that in some workflows AI should not enter at all.

That is the AI Maestro work: not a prefabricated framework, but the diligence Satya would have wanted in 2022, sized for a company that does not bill $50B a year.

Two costs the flown-in FDE will never explain well:

Criterion debt. When you pull the human out of the loop, every silent decision where the model was wrong accrues a debt. A week later the board asks, “where did that number come from?” and you cannot answer. It charges compound interest. A Google FDE does not prevent it; its deliverable is production, not decision sovereignty.

Shadow AI. While leadership debates whether to “do AI,” employees are already uploading confidential documents to free ChatGPT accounts from their phones. Information leakage, personal data leaving the perimeter, contracts running through a public model with no governance. It is happening inside your company today. An OpenAI FDE does not turn it off; it just formalizes it later.

Three questions before signing with any AI partner

Before your next meeting with any vendor using “AI transformation” in a discovery email, ask three questions. They separate the real FDE from the salesperson with a new title.

The first is genuinely uncomfortable: show me a project where your Go/No-Go was No. If they have never told a client no, you have your answer.

The second measures diligence, not marketing. Ask what happens when the data is dirty. The correct answer is not “we clean it with AI”; it is “first we map it, then we decide if AI can use any of it.” Discovery before installation. The difference shows up in the first invoice.

The third is the one almost nobody asks. Find out who will be in the chair six months from now when the board asks why an agent made a decision you do not understand. If the answer is “the FDE already rolled to the next client,” you are buying criterion debt packaged as innovation.

The most expensive rebrand of the decade is not consultants changing labels. It is the one sold when someone with a Google Cloud business card tells you “we’ll send an FDE” knowing the FDE, in Central America, is never going to land. You either build the thin layer yourself or pay 3x for it two years later, when a Big Four practice spends its first six months interviewing your team to discover what your team already knew.

The AI Maestro residency continues past the Go/No-Go for that reason: month-to-month advisory on what shifted in the market and which of those shifts apply to your operation. The only way to avoid redoing the diligence every twelve months when the next rebrand drops.