Why did Google and Accenture answer "what comes next" with more people?
Enterprise software spent two decades getting easier to buy. AI promised to push that further. Google Cloud and Accenture just bet the other way, on sending people in.
Cloud logins replaced big installs. Subscriptions replaced giant purchases. Vendors now talk about agents running whole business processes. On September 8 the two companies launched the Accenture Gemini Enterprise Business Group. Google will train up to 1,000 Accenture forward-deployed engineers, or FDEs, who work on-site with customers to plan and build AI applications on the Gemini Enterprise platform. They will not go alone. Accenture says the teams also carry industry, process and data specialists who sit with internal IT and with functional leaders like the CFO.
Google Cloud framed the move around delivery, not model quality. CEO Thomas Kurian said the new group "significantly expands the expertise and resources available to help our customers deliver real business value." Accenture chair and CEO Julie Sweet put it in terms of outcomes: "The companies seeing the greatest outcomes from AI are unlocking new growth, increasing productivity and resilience." The unstated half is that plenty of companies are not there yet. For all the attention on smarter models, closing that gap is what the 1,000 engineers are for.
The model is not the company
Why is a capable model not enough on its own? Because a model is not a process. A chatbot can summarize a contract in seconds.
A contract-review system a multinational trusts is a different job. Someone has to decide which contracts the system can read, where those documents live and which version wins when two disagree. Someone has to map the approval chain, the exceptions and the people allowed to overrule the system. Someone has to know which market's rules apply. Then the output has to land inside whatever tool the staff already use to get the work done.
The model might be the most impressive part of that system. It can also be a small slice of the actual problem. Google has been circling this point for months. In April it said its integration partners already had more than 330,000 experts trained on Google AI, and it put a $750 million partner fund behind programs that embed engineers closer to customer operations. A company can buy access to an excellent model this afternoon. It cannot buy a clean picture of its own business that fast.
Why do real businesses turn out messier than the demo?
Most AI demos start with something the real world rarely hands you: a clean, well-defined problem. Inside a running company, that is where the work hides.
Here is the document. Here is the data. Here is the task. Each of those sentences can hide months of effort. Customer records might sit across Salesforce, an ERP system, email, spreadsheets and a homegrown database someone built twelve years ago. Two departments use the same word for different things. Permissions that look tidy on an org chart have collected years of exceptions. A process labeled "standard" might really depend on three veterans who know when the standard should be ignored.
IBM describes the same data problem in its work on context engineering for enterprise AI: company information is scattered across applications, clouds, warehouses and documents, and the definitions, governance and lineage differ from system to system. A model does not inherit the institutional knowledge that reconciles all of it. That is why the forward-deployed engineer matters. The role sits between software engineer, consultant, analyst and translator. The engineer learns enough about the customer's business to turn a fuzzy operational problem into something software can handle. It is a very human job hiding inside an automation story.
Free CPA AI Policy Checklist
Before staff paste client data into AI, check the rules your firm is missing.
Get the free checklist and join the Dispatch for practical AI controls, vendor questions, and client-data safeguards for accounting teams.
Free. No spam. You will also get the Nexairi Dispatch.
Is enterprise AI reversing the SaaS model?
Cloud software trained buyers to expect less setup labor over time, not more. One product, sold to thousands of companies. Forward-deployed AI bends that.
The old economics improved as the product got more standardized. Every new customer did not need another team of engineers moving into the building. Palantir made the embedded-engineer approach well known years ago. Now the largest AI vendors and cloud providers are borrowing pieces of it. The logic holds. If the bottleneck has moved from model capability to organizational rollout, putting strong technical people next to the customer is one of the fastest ways through. Nexairi covered the same shape when OpenAI built a standalone company just to deploy its enterprise AI.
The question is what happens after. If every major deployment needs a pod of skilled people to map processes, reconcile data and redesign workflows, then the services layer is not just helping sell the product. It is becoming part of the product. That is why old-line consulting firms suddenly look central to an industry that spent years selling automation. Accenture has more than 750,000 people and decades of practice untangling the exact organizational mess AI vendors keep hitting. Google brings the models and the infrastructure. Neither side replaces the other.
What the "services as product" shift means for buyers
If the embedded team is part of the product, then switching costs are not just licenses. They are the workflows, data pipelines and quality checks the FDEs build around one vendor's stack. That is a real form of lock-in, and it is worth naming before the first team arrives. It also changes the comparison. The fair test is not model against model. It is total delivered outcome against total delivered cost, humans included. A weaker model with a team that understands your close process can beat a stronger model dropped in cold.
Does this mean enterprise AI does not work?
It is tempting to read 1,000 engineers as proof the hype failed. That is too simple. Big technologies have always needed people to fit them to the organizations using them.
ERP rollouts built enormous consulting businesses. Cloud migration produced another wave of systems integrators. Companies still staff large teams around software that matured decades ago. AI looks like the same pattern. The twist is that the enterprise AI pitch leaned hard on speed: fast understanding, fast answers, less human effort. The technology may do that once it has the right context. Getting the context in is turning into its own industry. It also explains why pilots look better than full rollouts. A pilot gets one clean problem, willing staff and vendor attention. Scale means dealing with every exception the demo skipped.
What should you ask before you buy?
A year ago the buying questions were about the model. Which one is smartest. What does it cost. The better questions now sit closer to the business.
Those model questions still matter. They just stopped being the ones that decide whether a rollout works. It is the same reason the seat count your finance team reports is already the wrong metric for AI value.
| Old question (about the model) | Better question (about the business) |
|---|---|
| Which model is smartest? | Which process are we changing, and who can explain its exceptions? |
| How large is the context window? | What data does the system need, and who controls that data? |
| What does the license cost? | If six skilled people sit inside the company for a year, is that in the ROI? |
| Which platform has the best agents? | Who owns the process after the implementation team leaves? |
None of this makes the spend irrational. It makes the real cost easier to see. A capable model sitting on top of a process nobody has mapped does not produce a capable business.
The human layer did not vanish. It moved.
There is something odd here. The technology built to cut human effort is creating steady demand for people who explain software and organizations to each other.
That may not be a phase. Companies are piles of history, politics, undocumented knowledge and old systems. Models can get better at reasoning over all of it. First someone has to make the place legible enough to reason about.
So the near-term winners in enterprise AI may not be whoever ships the smartest model. They may be whoever gets good at the slow, unglamorous work of wiring that intelligence into the mess underneath. Google and Accenture's thousand-engineer bet is a useful tell. The AI may automate a lot of work eventually. For now, humans are still being sent in to explain how the place actually runs.
Sources
- Accenture Newsroom: Accenture and Google Cloud Deepen Partnership with Formation of New Accenture Gemini Enterprise Business Group (September 8, 2026)
- Wall Street Journal: Google Cloud, Accenture Launch Unit to Put AI Engineers On-Site With Customers (September 8, 2026)
- PYMNTS: Google Helps Accenture Embed AI Engineers With Customers
- Google Cloud: Google Cloud Commits $750 Million to Accelerate Partners' Agentic AI Development (April 22, 2026)
- IBM: Context engineering for enterprise AI
Related Articles on Nexairi
Free Assessment
Is your firm ready for AI?
A 5-minute governance check for CPA firms using ChatGPT, Copilot or AI accounting software. Get your score and your top gaps — free.
Jim Smart is the founder and editor in chief of Nexairi. A Business Intelligence Developer with experience building data systems for Verizon, U.S. Army operations, and enterprise finance teams, Jim spent years turning complex data into decisions that executives could act on — dashboards, forecasting models, and automation pipelines across telecom and government contracting. He founded Nexairi to apply that same clarity to AI: making emerging technology understandable and actionable for the operators, accountants, and business owners who need it most. Jim holds GenAI certifications from the University of South Florida Bellini College of AI and completed Springboard's Data Science Career Track.
