7,000 Forward-Deployed Engineers in 68 Days: The New Enterprise AI Model

TL;DR

On 8 September Accenture and Google Cloud formed the Accenture Gemini Enterprise Business Group, with a stated 1,000 forward deployed engineers.

On 2 July Microsoft launched Microsoft Frontier Company, committing $2.5 billion and 6,000 people to embed AI engineers inside customers.

That is 7,000 embedded engineers committed in 68 days, by two of the largest vendors in enterprise software.

Divide Microsoft's own numbers and the commitment is about $417,000 per engineer. Against the roughly 9,000 clients Accenture reports serving, its 1,000 engineers work out at 1 per 9 of them.

The demand-side reason is measured, not guessed. Roughly 75% of organisations are adopting agentic AI, 17% have deployed agents, and 11% have production-ready agentic systems. Selling the software is not producing the outcome.

What was announced yesterday?

A joint unit, with a headcount and a job description.

Accenture and Google Cloud announced the Accenture Gemini Enterprise Business Group on 8 September, described as "a global group designed to help clients scale Gemini Enterprise outcomes in the agentic AI era." It sits inside the existing Accenture Google Business Group, and it commits 1,000 forward deployed engineers.

Julie Sweet framed it around outcomes: "The companies seeing the greatest outcomes from AI are unlocking new growth, increasing productivity and resilience." Thomas Kurian was blunter about the gap being addressed: "Deploying agentic AI is a top priority for enterprises today, and the Accenture Gemini Enterprise Business Group significantly expands the expertise and resources available to help our customers deliver real business value."

Read that second quote carefully. The constraint it names is not the model, the cloud or the licence. It is expertise and resources at the customer's end.

Why does this look familiar?

Because Microsoft did the same thing ten weeks earlier, larger.

On 2 July, Microsoft launched Microsoft Frontier Company, committing $2.5 billion and 6,000 employees to embed AI engineering talent directly inside enterprise customers. Its stated mission was to "co-design, co-innovate, deploy and continuously improve" large-scale AI systems inside customer operations, and it is led by Rodrigo Kede Lima. Microsoft also guarantees intellectual property protection on jointly built systems.

Put the two side by side and the shape is identical.

Microsoft Frontier Company. Accenture Gemini Enterprise Business Group

Announced. 2 July 2026. 8 September 2026

Embedded engineers. 6,000. 1,000

Money stated. $2.5 billion. Not disclosed

Model. Engineers placed inside the customer. Forward deployed engineers, plus training and certification

Named problem. AI systems that do not reach production. Scaling agentic outcomes

7,000 embedded engineers, committed in 68 days. That is roughly 103 engineers a day of announced capacity, from two vendors, aimed at the same bottleneck.

What do the numbers say when you divide them?

Two divisions are worth doing, because both announcements are quoted as headline totals and neither is usually scaled.

Microsoft's cost per engineer. $2.5 billion across 6,000 people is about $417,000 each. Microsoft has not said over what period, so read that as the size of a programme rather than a salary. Even so, it sets a floor on what the company thinks an embedded AI engineer is worth committing.

Accenture's coverage per client. Accenture reports approximately 9,000 clients and around 799,000 employees. So:

1,000 FDEs / 9,000 clients = 0.11 engineers per client

1,000 FDEs / 799,000 employees = 0.125% of headcount

1,000 FDEs / ~50,000 Google Cloud-skilled staff = 2%

One forward deployed engineer for every nine clients. That is the honest scale of this. It is a programme for a concentrated set of large accounts, not a capability that arrives with your subscription. If you are reading the announcement wondering whether it changes your project, the arithmetic says: only if you are already in the top tier of the account list.

What is the demand-side reason?

This is the part the announcements do not say out loud, and it is measurable.

Three independent surveys, taken across 2026, describe the same gap:

Source. Finding

Forrester, June 2026. About 75% of organisations are adopting agentic AI

Gartner CIO Survey 2026. Only 17% have actually deployed agents

Deloitte Tech Trends 2026. Only 11% have production-ready agentic systems

intent 75%

in production 11% to 17%

the gap 58 to 64 percentage points

Between 58 and 64 points of stated intent are not reaching production. Seven thousand engineers are being pointed at that gap. That is what "forward deployed" means in practice: the vendor concluded the software alone was not closing it.

There is a sharper version of the same signal in McKinsey's State of AI 2026, published 25 August from 1,719 responses across 97 countries. 32% of organisations decided against buying off-the-shelf software because agentic coding tools could build it instead. By sector: technology 41%, healthcare payers and providers 39%, professional services and energy and materials 38% each, financial institutions 36%. Among the high performers, the 6% attributing at least 5% of EBIT to AI, nearly half skipped a purchase against 31% of everyone else.

Set those two findings next to each other and the strategy reads clearly. A third of buyers now think they can build it, and only about an eighth of adopters can actually run it. Both statements are about the same missing thing, which is engineering capacity at the customer. One vendor response is to sell more software. The response both Microsoft and Google chose is to supply the engineers.

Is this a good deal for the buyer?

Partly, and the caveats are worth naming before signing anything.

What is genuinely good. An engineer inside your operation sees your data, your exceptions and your edge cases. That is where AI projects actually fail, and no amount of platform capability substitutes for it. Microsoft's IP guarantee on co-built systems is a real term, not a slogan, and it is the right question to ask of any equivalent arrangement.

What to watch. A forward deployed engineer employed by the platform vendor has one structural bias: the answer will involve the platform. That is not dishonesty, it is org chart. When we compared how four vendors govern agents, the differences were real and consequential, and they are not differences an embedded engineer from any one of them is well placed to weigh.

What it does not change. Yesterday we covered Business Central moving Copilot extension AI onto Microsoft-managed resources, which removes infrastructure from the partner and puts consumption on the customer's meter. The forward-deployed model does the same thing with people. Both trades are real in both directions: less to run, and less that you see.

What should a buyer do about it?

Ask whether you are in scope before you plan around it. One engineer per nine clients is a rationing ratio. Get the answer in writing rather than inferring it from a press release.

Separate the two questions the gap contains. "Can we build it" and "can we run it" are not the same, and the survey data says organisations are more confident about the first than the second. A build decision made on coding-agent speed still inherits the run cost.

Price the run, not the build. McKinsey's 32% describes purchases avoided. It does not describe systems successfully operated a year later. That number does not exist yet, and its absence is the most important thing about the finding.

Keep one person who is not the vendor's. The value of an embedded engineer goes up, not down, when somebody on your side can weigh their recommendation against an alternative.

FAQ

What is a forward deployed engineer?

An engineer employed by the vendor but working inside the customer's organisation, on the customer's systems and problems, rather than delivering a defined project from outside.

Is the Accenture group new headcount or reassigned staff?

The announcement states a 1,000-person forward deployed engineer workforce and says Google Cloud will train them. It does not state how many are net new.

How big is Microsoft's commitment in context?

$2.5 billion and 6,000 employees, described in the reporting as the largest single commitment to forward-deployed AI engineering from a major software vendor to date.

Do these numbers overlap?

No. They are separate programmes at separate companies, which is what makes the convergence interesting.

Does this mean agentic AI is failing?

It means deployment is the constraint rather than capability. Roughly 75% adopting against 11% to 17% in production is a delivery gap, not a technology verdict.

The last mile

Two of the largest vendors in enterprise software looked at the same data and reached the same conclusion inside ten weeks: the product is not the bottleneck, the deployment is.

That is an unusually candid admission to make with a chequebook. It is also, read from the buyer's side, an argument that the scarce thing is not access to models or platforms but people who can connect them to how a business actually runs. At one engineer per nine clients, most organisations will not receive that from the vendor. The alternative is not a bigger licence. It is having the capability close enough to your own systems that the exceptions, the edge cases and the run cost are somebody's actual job. That is the layer Cognilium builds in, and this week two very large companies spent a great deal of money agreeing it is where the work is.