What does Wonderful actually sell?

Wonderful sells AI agents for customer service and back-office work, delivered by engineering teams that work inside the customer's environment and connect the agents to the customer's own systems. Its founder Bar Winkler told Globes in June 2026 that the product on its own is worth little until it is connected to dozens of systems inside the organisation, so the delivery work is the product as much as the software is.

Why did Wonderful grow so fast when many AI agent startups did not?

By its own account and by the reporting of CTech and Globes, it won named enterprise customers within its first four months and led every later case study with a time-to-production figure. That gave each funding round a concrete thing to underwrite: a working system inside a real bank or telco, with the customer's executive quoted on the result.

Is Wonderful's business a software company or a services company?

CTech reported in September 2026 that its gross margin is 52 percent, against 70 to 90 percent for software companies, and that 400 of its 650 employees work at customer sites. The open question CTech raised is whether that cost structure becomes software economics over time or stays a services business with an AI product attached.

What is the lesson for a company buying AI?

Budget for the last mile, because that is where the money went. The model is available to everyone at the same price, so the cost and the risk sit in connecting it to your billing, identity and ticketing systems, checking that it behaves, and changing how your people work around it.

How did Wonderful get its first customers?

CTech reported in July 2025 that the company built a 50-person engineering team to make agents that work in Hebrew, then approached large Israeli companies directly, winning Maccabi Health Services and the telco Bezeq. Sales conversations were under way within three and a half months of starting, which is the pace a new company has to move at when it has no history to point to.

Does an AI vendor need a local team in every country?

Wonderful's answer is yes for large enterprises, and Winkler told Globes that working with the largest organisations needs a presence on the ground wherever you operate. The reason is practical: the integration work, the regulatory questions and the change to daily routines all happen inside the customer's building, so someone has to be in it.

What should a buyer ask a vendor whose model is forward deployed engineers?

Ask whether the second deployment cost less than the first, and whether your own people shipped anything without the vendor after the handover. If every engagement starts from zero, the venture capital analyst Eze Vidra wrote in September 2026, the company is building a consultancy, and the same test applies to the price you will pay in year two.

What did the three funding rounds in one year reward?

Each round followed a visible jump in production use. Globes reported annual recurring revenue of 1 million dollars in August 2025, 7 million at the end of 2025 and 18 million in the first quarter of 2026, and the Series B and Series C came four and ten months after the Series A.

Are Wonderful's customer results independently verified?

No. The time-to-production figures and resolution rates come from Wonderful's own case studies, with customer executives quoted on Wonderful's site, so they are the company's account of its own work. The funding amounts, valuation, headcount and margin figures come from TechCrunch, CTech and Globes and are the independently reported part.

Wonderful went from zero to a $5 billion valuation in 20 months. Here is what it proves about enterprise AI: Reveneau

Q: How does Reveneau's approach differ from Wonderful's?

We do the same last-mile work, taking a system from specification to production inside the client's stack and owning it after release, but AI writes the code and an eval suite written from the specification proves every change before it ships. The point of that method is to make the second deployment cheaper than the first without adding people, and we state it as the mechanism we work by, because no saving has been measured yet.

For the past two weeks we have read everything published about Wonderful, the company founded in Tel Aviv in early 2025 that closed a $550 million round at a $5 billion valuation on 2 September 2026, as TechCrunch reported. We wanted to know what a company that young did to get there, because it sells something close to what we sell: getting AI out of a demo and into production inside a real business.

The short version is that every dollar went to production. Here is what we keep seeing in its 20 months: buyers paid for production and only for production, the proof that opened the money was a named customer won in the first four months, and doing the last mile with people costs the margin, in public, in a way every buyer can now read.

1. The buyers paid for production, and only for production

Wonderful started as voice agents that speak Hebrew, and within a year it was selling agents for every channel a customer uses. Its founder Bar Winkler told Globes in August 2026 that "no company wanted just voice agents": customers wanted agents in chat, in email, in sales and in the back office, wherever an employee moves information from one system to another. What they were buying was the connection to their own systems, and the product followed the connection.

Winkler said the same thing more bluntly in a June 2026 interview: "The product itself isn't worth much if you don't connect it to dozens of systems within the enterprise", and, of the work his company does, "We need to connect systems, understand processes and persuade organizations to change the way they work." The company's own site now leads its deployment page with the line "Deployment is where enterprise AI succeeds or fails."

The mechanism is simple. Every bank can call the same model from the same vendors at the same price. The scarce thing is a system that reads the bank's billing records, respects its permissions, survives its security review, and behaves the way the bank's own agents are trained to behave. That is engineering inside the customer's building, and the customer will pay for it because it is the only part they cannot buy off a price list.

So if you are the one buying AI, the lesson is where to put the budget. The model line in the estimate will be the smallest one. The integration, the checks, and the change to how your people work are the build. We wrote about this pattern before Wonderful's numbers made it visible, in An AI demo is not a product, and the numbers have since done the arguing for us.

2. The proof was one named customer, and it arrived in four months

CTech's report on the seed round, on 2 July 2025, is the most useful document in the whole story, because it describes the company before the money arrived. Winkler had built "a strong engineering team of 50 employees to develop AI agents that can operate in Hebrew", approached large Israeli companies directly, and won Maccabi Health Services and the telco Bezeq. Sales conversations were running within three and a half months of starting.

That is the whole method. A company with no history cannot show a track record, so it showed a customer instead. The seed announcement carried a quote from Bezeq's chief executive saying the company had evaluated over a dozen AI solutions, by Wonderful's own account of the round. Four months later came a $100 million Series A, reported by TechCrunch. Four months after that, $150 million at a $2 billion valuation, again per TechCrunch.

The revenue line the rounds were following is public too. Globes reported annual recurring revenue, the yearly value of the subscriptions the company had signed, of $1 million in August 2025, $7 million at the end of 2025 and $18 million in the first quarter of 2026, and CTech put the run rate at $70 million on the day of the Series C. Each round followed a jump in production use.

What's changed since the first customer is how the proof is written. Every Wonderful case study now opens with a clock: 72 hours to a first voice agent at Bank Hapoalim, 19 days to production at the Colombian bank Banco Caja Social, five weeks at the Greek postal service ELTA. These are the company's own accounts, with the customer's executive quoted on Wonderful's site, so read them as claims. But notice the shape of the claim. It is a date the contract was signed and a date real users were served, and the gap between them is the number.

For a reader choosing a partner, that is the question to carry into every conversation: when was the contract signed, and when did the first real customer get served? A vendor that has done it will answer with two dates. One that has not will answer with a demo.

3. Doing the last mile with people costs the margin, and everyone can see it

Wonderful pays for production with headcount. Globes reported that the company went from 90 employees to around 630 in one year. CTech's Sophie Shulman reported on 6 September 2026 that about 400 of its 650 employees work at customer sites, and that its gross margin, the share of each dollar of revenue left after the cost of delivering the work, is about 52 percent, against the 70 to 90 percent a software company usually keeps.

That is the bill for the last mile, printed in a newspaper. Shulman also reported that the company can grow this fast partly because it offers its services at low prices compared with Salesforce and Accenture, and drew a parallel to Uber's early years, when venture capital subsidised prices before the company had to face the economics of operating at scale. Wonderful has raised more than $800 million, per CTech, so it can afford to subsidise deployment for a long time.

Whether it should is the argument the investors are having in public. Andreessen Horowitz's Joe Schmidt made the case for it in June 2025 in an essay titled Trading Margin for Moat: ServiceNow had a gross margin of 63.2 percent at its IPO and reached 79 percent later, Workday started at 54.1 percent and reached 75 percent, and the implementation work that cost the margin early is what made the software hard to remove. The venture capital analyst Eze Vidra wrote the counter-test on 3 September 2026: "fieldwork only becomes a moat when it becomes product. If every engagement starts from zero, the startup is building a consultancy."

That test is the one we would apply to ourselves, and to any vendor whose model is engineers at your desk. The last mile is real work and someone pays for it. There are only three ways: the vendor pays out of its margin, the customer pays with its own people after a handover, or the work itself gets cheaper each time it is done. The first two are how Wonderful is paying today. The third is the only one that scales without hiring 540 people in a year.

The third way is also our whole method. We do the same last-mile work Wonderful does, from specification to production inside the client's own stack, and we own it after release. But AI writes the code, and an eval suite written from the specification proves every change before it ships, so the second deployment should cost less than the first without adding people. We say "should" on purpose. We have not measured that number yet, and our pages label every timeline as the one we scope to. When we have the measured figure we will print it, and until then we would rather be the vendor that says so.

Our thanks to the reporters at CTech, Globes and TechCrunch, whose work this piece rests on, and to Wonderful's team for publishing enough of its own thinking to be read closely. Everyone can buy the model. Nobody can buy production.

Sources

TechCrunch, 2 September 2026: the $550 million Series C at a $5 billion valuation, led by Insight Partners with Salesforce joining.
TechCrunch, 12 March 2026: the $150 million Series B at a $2 billion valuation.
TechCrunch, 11 November 2025: the $100 million Series A.
CTech, 2 July 2025: the seed round, the 50-person Hebrew-first team, and the first customers Maccabi and Bezeq.
CTech, 2 September 2026: the $70 million run rate, 650 employees, and more than $800 million raised.
CTech, Sophie Shulman, 6 September 2026: the 52 percent gross margin, 400 of 650 employees at customer sites, pricing against Salesforce and Accenture, and the Uber parallel.
Globes, 17 June 2026: Bar Winkler's interview and the revenue sequence from August 2025 to the first quarter of 2026.
Globes, 12 August 2026: headcount from 90 to around 630 in a year, and the "no company wanted just voice agents" quote.
Index Ventures, 2 July 2025: the seed announcement carrying the Bezeq quote, which is the company's and its investor's own account.
Wonderful, Banco Caja Social case study: the 19-day figure, a company claim.
Andreessen Horowitz, Joe Schmidt, 4 June 2025: Trading Margin for Moat, with the ServiceNow and Workday gross margin figures.
VC Cafe, Eze Vidra, 3 September 2026: the "fieldwork only becomes a moat when it becomes product" test.