By SavantCat
Last week I was showing someone the shop-floor AI customer service demo I built, and a business owner asked: “Is what you make the same thing as those large model companies do?”
At the time I answered vaguely. Only in the past few days, reading about the funding round of an Israeli company called Wonderful, did I think that question through.
On 2 September 2026 Wonderful announced a USD 550 million Series C, taking its valuation to USD 5 billion, about RMB 33.5 billion. The company is not yet 20 months old. Its youngest co-founder and CTO is 24. Headcount grew from 350 at the start of the year to 650, and annualised revenue went from about USD 1 million to about USD 70 million — a 70-fold rise in half a year.
Capital prices that story at RMB 33.5 billion. But what really hit me is the way it makes money:
It did not train a new large model, and it did not chase the all-purpose AI assistant bandwagon. It does the least glamorous, least sexy thing in AI — sending people on site to help enterprises actually use AI.
This piece unpacks three things: why this company is worth RMB 33.5 billion, where AI money is actually flowing, and what it means for people like us who do not build large models and just want to earn a living with AI.
1. It chose a cut that looked harder
Wonderful did not start by attacking the most crowded English-language market. It chose AI customer service for non-English markets.
Why customer service? Because it is naturally high-frequency, has clear rules and produces measurable results — the place where AI shows effect fastest. But it is also extremely dependent on language and culture: Italian customers like warmth, Dutch customers want directness, Japanese customers expect indirectness. Wonderful calls this cultural fluency: an AI agent must adjust tone, wording and even the breathing rhythm of a conversation for each market.
But customer service is only its entry point, not the destination. Once a service agent is inside an enterprise it has to connect to the CRM, to orders, to payments, to internal workflows. It is no longer a bot that answers your messages; it gradually becomes a node inside the enterprise's business systems that actually does work.
So the company redefined itself too: from an AI customer service company to an enterprise-grade AI operating system.
2. What makes it valuable is exactly the link AI most lacks
Over the past two years the barrier to enterprises buying AI has fallen very low. Chatbots, knowledge bases, meeting notes, coding assistants — ready-made products are everywhere, and buying any one of them completes a first round of AI transformation.
The genuinely hard part is step two: getting AI into an enterprise's real business processes.
A service agent has to call your order and payment systems; a sales agent has to know your customer history and internal permissions; a finance agent has to connect to your ERP and approval flows. Then you realise: models are worth less and less, while the people who can embed models into enterprise processes are worth more and more.
Wonderful's sharpest move is something called FDE, forward-deployed engineers. Unlike traditional software companies that sell and walk away, it sends its own engineers to live inside the client's company, working with the client's team to take the first AI application from design through integration to launch, running it in production for real, and handing it back only once it works.
In plain language: software itself is increasingly easy to buy; what is truly scarce is people who can install software in your company and make it run reliably.
The playbook is not new — the big data company Palantir built itself from a mocked software company into a giant worth hundreds of billions by putting engineers on site. But now it has become a direction the whole AI industry is chasing:
- AWS set up a dedicated forward-deployed engineering organisation in June this year, committing USD 1 billion to deploy thousands of engineers;
- Google Cloud put USD 750 million into agentic AI in April, including helping partners build FDE teams;
- Anthropic joined with Blackstone and other institutions in May to set up an independent enterprise AI services company, putting Claude into the core business of mid-sized enterprises;
- Even Salesforce broke with precedent and invested in Wonderful this time.
Cloud providers and model vendors have finally worked something out: the bottleneck in enterprise AI commercialisation has long since stopped being model capability and become who can deploy models into enterprises and make them run.
3. What does this mean for those of us who do not build models?
Reading this story, my back felt a little cold and a little warm at once.
Cold, because if even AWS and Google have to send engineers on site to install AI for customers, then this road is heavy work, dirty work, work that needs people sitting with a client grinding at it. That is a world away from my original idea of writing a tool and earning passively.
Warm, because this is exactly the window of opportunity for ordinary people.
Capital is willing to price helping enterprises use AI at USD 5 billion because they have understood it — models will keep getting cheaper and more commoditised, but delivery will always need people who understand the business and are willing to be on site. The big firms have tens of billions, but they cannot reach and cannot be bothered with the work of the hardware shop downstairs, the health check centre, or the bookkeeper who still issues invoices by hand.
The shop-floor AI customer service I built is essentially a small-business version of the Wonderful story: not selling a smarter bot, but helping an owner catch the enquiries missed at midnight, remember the old orders it took three searches to find, and file the questions that could not be answered into categories. It is not sexy, but owners understand it — one fewer person staying up to answer the phone.
What it comes down to for you and me — three sentences
First, do not be dazzled by building large models; earn money from helping people use them. Models are the giants' game; delivery is everyone's opportunity. Wonderful proved with a USD 5 billion valuation that installing AI into a concrete industry is a big business validated by global capital.
Second, pick one role with clear processes and deep pain. Customer service, reconciliation, tender documents, order review, after-sales — wherever the rules are rigid, repetition is high and people used to watch it, that is where AI slots in most easily and where owners are most willing to pay. Do not be greedy; make one point work thoroughly first.
Third, do not wait for a stronger model before you start. Technology will keep changing, but the need to help an owner get work done well and keep the accounts straight will not. The AI available today is already enough to serve one small shop — use it first, get the first one working, then talk about scaling.
The first half of AI was geniuses on stage talking about how big the model is and how many parameters it has. The second half will leave the stage to people willing to sit on site with a client and install AI into a real business piece by piece. I do not know how many billions the former is worth; the latter is already being priced by global capital.
Is there a job near you with very rigid rules and heavy manual effort? Or are you yourself doing something like helping a small shop use AI? Talk about it in the comments — the next opening may be hiding in these unremarkable corners.
#AIDelivery #EnterpriseAI #AIAgents #SmallBusiness #AIMoney