Credit: Accenture The chief executive of a consultancy with 799,000 staff has described what her clients are telling her about AI, and it is not a sales pitch. “We promised AI can do that,” Julie Sweet told the Wall Street Journal. “And they’re like, we get it, except it’s not happening, help us make it happen.” Sweet was explaining why Accenture and Google Cloud have formed a new unit. Isabelle Bousquette reported the plan for the Wall Street Journal an hour before the companies announced it. What they are actually building The Accenture Gemini Enterprise Business Group will put engineers inside client offices. Google Cloud will help train up to 1,000 Accenture forward-deployed engineers, who work on-site to plan and build applications on Google’s Gemini Enterprise platform.
Accenture industry and process specialists sit alongside them, plus a small number of Google Cloud engineers reserved for top-tier customers. Neither company would disclose financials. Both called it a significant investment. The model is not new.
Palantir popularised the forward-deployed engineer, and the term has since escaped into general use. Everyone has reached the same conclusion This is the fifth version of the same idea we have covered this year, and the consistency is the story. Microsoft launched Frontier in July, committing $2.5bn and 6,000 engineers to enterprise AI deployment, two days after Amazon Web Services announced a billion-dollar equivalent. Anthropic and Blackstone built Ode, a $1.5bn venture selling implementation rather than models.
OpenAI set up a standalone services firm. TCS said it would convert up to 1.5% of its workforce into 8,900 deployment engineers. Five companies, one answer. When the technology does not deliver, send people.
What that says about the product Software was supposed to be the thing that scales without adding humans. A platform requiring 1,000 trained engineers per partnership is not behaving like software. It is behaving like a project, and projects carry consultancy economics rather than software economics. Sweet described how Accenture already works.
A pod goes into a company containing engineers, process people and data specialists. With one client it started on invoice processing. Over eight to 12 weeks the pod mapped the process, integrated the data, built an agentic system and wrote a plan for scaling it. Then it handed the work to other Accenture staff to actually scale.
Eight to 12 weeks, for invoice processing, at one company. That is the unit of work. Accenture is on both sides Here is the detail worth pausing on. Accenture is already a named partner in Microsoft Frontier.
It is now the anchor partner in Google Cloud’s version. It has nearly 50,000 Google Cloud-skilled staff, and it presumably has comparable numbers trained on Microsoft. That is not disloyalty. It is the business, and it tells you where the scarcity sits.
The models are abundant and increasingly interchangeable. The people who can install them inside a bank are not. Which is why the platform companies are paying a consultancy to train its own staff on their software, rather than the other way round. The reference customer is in-house The announcement offers one named success story.
YouTube deployed a Gemini Enterprise agent during a demand surge for NFL Sunday Ticket, and the companies say customer sentiment rose 11% while average handling time fell 37%. Those are good numbers. YouTube is also owned by Alphabet, which owns Google Cloud. So the flagship proof that Google’s enterprise platform works comes from Google deploying it on itself, with Accenture’s help.
That does not make the result false. It does mean the first external case study is still outstanding. The seller has the same problem Accenture is not selling this from a position of strength. Its own shares have been marked down on a worsening outlook and an unclear AI future, which is to say the market applied to Accenture the exact doubt Accenture is now offering to resolve for others.
Sweet’s answer is experience: “You have to know: how do you actually deploy a system in a big company?” That is a reasonable defence and it is also the whole bet. If deployment expertise is the bottleneck, the firms with 799,000 people and fifty years of enterprise scar tissue win. If the models get good enough to deploy themselves, those firms are selling scaffolding for a building that no longer needs it. Cognizant showed how uncomfortable that position is last week.
It published research putting $4.5tn of US wages in the path of AI, and hired 1,500 graduates the same day. Europe should watch the platform, not the unit Gemini Enterprise is the thing being sold here, and it has been arriving in regulated European sectors already. Google launched Gemini Enterprise for Legal in August with Freshfields, Cleary and Weil, connecting into iManage, NetDocuments and Thomson Reuters. A thousand consultants trained to install that platform is how it reaches European enterprises at scale, and how it becomes the default rather than one option among several.
The DMA governs how Google behaves in search. Nothing comparable governs how an enterprise AI platform becomes embedded in a company’s finance function. What to watch Whether the 1,000 engineers are hired or retrained. The Journal reports Google will help train up to 1,000, and the announcement describes establishing a 1,000-person workforce.
Those are not quite the same, and the difference is whether this is net new employment. Whether a named external customer appears with numbers as specific as YouTube’s. And whether any of these deployment businesses report revenue separately. Microsoft put $2.5bn against Frontier and TCS put its AI business at $2.6bn annualised.
Neither Accenture nor Google Cloud has put a figure on this one. Also tagged with
