Who founded Wonderful and when?

Bar Winkler and Roey Lalazar founded it in Tel Aviv in early 2025, according to CTech's report on the seed round in July 2025. Winkler had founded Approve.com and sold it to Tipalti in 2021 and was an early employee at ironSource, and Lalazar had founded Kaps, a localisation platform, so the pairing was a payments operator and a language technologist.

How much money has Wonderful raised and at what valuations?

A $34 million seed in July 2025, a $100 million Series A in November 2025 at a reported $700 million valuation, a $150 million Series B in March 2026 at $2 billion, and a $550 million Series C in September 2026 at $5 billion, with a further $170 million secondary sale by employees and early angels. CTech puts the total raised at more than $800 million in under two years.

What is Wonderful's revenue?

Globes reported annual recurring revenue of $1 million in August 2025, $7 million at the end of 2025 and $18 million in the first quarter of 2026, and on the day of the Series C CTech put the run rate at an estimated $70 million, expected to pass $100 million by the end of 2026. None of these figures is audited or disclosed by the company itself.

What does Wonderful sell today?

It now calls its product an AI operating system for the enterprise: agents for customers and employees, an agent builder, a gateway that routes requests between models, and evaluation and governance tools, delivered by engineers who work inside the customer's environment. It started in 2025 as voice agents for customer service in non-English languages and renamed its category three times in 14 months.

Why does Wonderful put engineers inside customer companies?

Because, in its founder's words to Globes, the product on its own is worth little until it is connected to dozens of systems inside the organisation, and that connection work happens in the customer's building. CTech reported that about 400 of the company's 650 employees work at customer sites, and the company's careers page lists forward deployed engineer and deployment strategist as its two main technical roles.

What is Wonderful's gross margin and why does it matter?

CTech reported in September 2026, citing the Wall Street Journal, a gross margin of about 52 percent, against the 70 to 90 percent a software company usually keeps. Gross margin is what is left of each dollar of revenue after paying to deliver the work, so the gap is the cost of the deployment teams, and whether it closes is the main question analysts ask about the company.

Which customers has Wonderful named?

Independently reported customers include Bezeq, Maccabi Health Services, Bank Hapoalim, Bank Leumi, Israel Discount Bank and the Israel Electric Corporation in Israel, and FC Bayern in Germany. Its own case studies name OTE in Greece, Banco Caja Social in Colombia, Petrol Ofisi in Turkey, Telefónica's Colombian operation, PPC Energie in Romania, ELTA in Greece and Mercado Libre's vehicle marketplace in Mexico, and the results in those studies are the company's own account.

What do investors say about Wonderful?

Index Ventures, which led the seed and Series A, framed the bet as reaching what it called AI deserts where US-centric products do not work, and Insight Partners, which led the Series B and Series C, wrote that the next wave of adoption will come from local depth. Salesforce joined the Series C as a new investor, which TechCrunch noted was its first investment in the company.

What do critics say about Wonderful?

The main criticism is about the economics. VC Cafe's Eze Vidra wrote that fieldwork only becomes a moat when it becomes product and that a startup whose every engagement starts from zero is a consultancy, and a 3V analysis noted the company has not disclosed gross margin, contract value or the share of revenue that is deployment work, and that its category vocabulary shifted to AI operating system just before the Series C.

Did Wonderful actually ban manual coding?

Its chief architect Daniel Sikorskiy wrote on 10 March 2026 that the company banned hand-written code four months earlier, with no phased rollout. The post describes building infrastructure that made models test their own work before a task counted as complete, and claims the Agent Builder product was roughly 90,000 lines of code built in about two weeks, a figure nobody outside the company has checked.

What inconsistencies did you find in Wonderful's public record?

Several small ones: its Germany page names one general manager and its about page names another, Globes headlined the Series C at $5.5 billion while the company and TechCrunch say $5 billion, the Series B headcount was 300 in TechCrunch and 350 in the press release, and on the day of the Series A the founder said 10 countries while the lead investor's post said 25 or more and 30. None changes the story, and all are what a company growing this fast looks like from outside.

What should another company take from Wonderful's story?

That buyers paid for production and never for a model, that the proof which opened the funding was a named customer won in the first four months, and that doing the last mile with people has a visible price. The open question, whether the second deployment gets cheaper than the first, is the one any vendor with engineers in a customer's building should be able to answer, including us.

The Wonderful story: how a 50-person Hebrew voice-agent team became a $5 billion company in 20 months, and what everyone says about it: Reveneau

We spent two weeks reading everything published about Wonderful: 27 press items on its own site, the funding coverage in TechCrunch, CTech, Globes and Reuters, its investors' essays, its own founders' posts, its country pages, its careers page, the analyst pieces that followed the Series C, and the handful of places where employees and customers say anything at all. This is the fullest account we could assemble. Every figure carries its source, and every number that comes from the company or its investors is labelled as the company's own account, because a company's description of itself is a claim and never a measurement.

We wrote it because Wonderful sells something close to what we sell, getting AI out of a demo and into production inside a real business, and because its 20 months are the clearest public record yet of what that work is worth and what it costs. Here is the story, then what everyone says about it, then the loose ends.

Part one: two founders and a thesis about Hebrew

Wonderful was founded in Tel Aviv in early 2025 by Bar Winkler, its chief executive, and Roey Lalazar, its chief technology officer, according to CTech's report on the seed round. Winkler had founded Approve.com, a payments platform, and sold it to Tipalti in 2021. Lalazar had founded Kaps, an AI localisation platform, which is the kind of background that makes the company's first bet look less strange than it sounds.

Index Ventures, which led the seed round, added colour in its own announcement that the press did not independently confirm: that Winkler was one of the earliest employees at ironSource and "founded and sold Approve.com for $40M within two years", and that Lalazar had bootstrapped a business to $1 million in revenue by age 22 and led a unit in Israeli military intelligence. Those are an investor's words about its own portfolio company. Hold them at that weight.

The first bet was a language. Winkler told CTech in July 2025: "We started working on Wonderful at the beginning of the year and built a strong engineering team of 50 employees to develop AI agents that can operate in Hebrew." Then: "We created a product that speaks naturally in Hebrew, with extensive functionality, and approached several large corporations in Israel." The company had raised $4 million from angels before the seed, by Winkler's account in the same interview.

The thesis behind the language was that the English-speaking market was crowded and the rest of the world was empty. Index's Hannah Seal later gave it a name in her Series A post, the hypothesis "that there are vast 'AI deserts'", places where, in her words, US-centric approaches do not work. Winkler's version, from the seed announcement: "We looked at the current state of AI and found it unbearable that most of the world will have to wait years for something that can be delivered today. Our strategy is simple: rapidly partner with the top enterprises in each market and build the talent density needed to obsess over delivering an amazing experience for their customers."

He also said, in that first CTech interview, something that reads differently now that the company sells everywhere: "When entering the U.S. market, you must stay focused: pick a specific vertical and compete there." Wonderful did the opposite by geography and the same by sequence. It picked one market, won it, and moved to the next.

Part two: the timeline, month by month

The dates below come from the press, the company's own blog index and its announcements. Where a figure is the company's own, the table says so.

Date	What happened	Source
Early 2025	Founded in Tel Aviv. 50 engineers building Hebrew voice agents. $4M from angels.	CTech, 2 Jul 2025
2 Jul 2025	$34M seed led by Index, with Bessemer and Vine. First customers Bezeq and Maccabi Health Services. Sales conversations within 3.5 months of starting.	CTech; Index
30 Sep 2025	Greece, "our first international office", per the company's Greece page.	wonderful.ai
Oct 2025	Netherlands and Switzerland launch posts. Amsterdam named as European headquarters.	wonderful.ai; Silicon Canals
11 Nov 2025	$100M Series A led by Index, with Insight, IVP, Bessemer and Vine. Reported $700M valuation. Winkler: "launched operations in 10 countries."	TechCrunch; CTech
12 Nov 2025	Poland launch, GM Marcin Motel.	ISBtech
7 Jan 2026	Winkler essay "The Execution Gap No Model Can Close."	wonderful.ai
23 Jan 2026	Agent Builder launched, built on Anthropic's Claude. Company claims 60+ deployments.	PR Newswire
10 Mar 2026	Chief architect's post "Going Codeless": manual coding banned four months earlier.	wonderful.ai
12 Mar 2026	$150M Series B led by Insight at $2B. Headcount 300 (TechCrunch) or 350 (release), target 900. 30+ countries.	TechCrunch; PR Newswire
6 Apr 2026	Alliance with McKinsey and QuantumBlack announced.	wonderful.ai
17 Jun 2026	Globes profile: ARR $1M Aug 2025, $7M end 2025, $18M Q1 2026; ~80 large customers; ~530 staff; 28 offices.	Globes
21 Jun 2026	"Computer use" announced: agents operate legacy systems through a screen on a virtual machine.	wonderful.ai
8 Aug 2026	India launch, Mumbai centre, plan to hire 1,000 engineers. Company says 100+ customers, 35+ markets.	wonderful.ai; ynet
12 Aug 2026	Globes: seeking $500M at $5B; ~630 staff, up from 90 a year earlier; ARR $55M.	Globes
2 Sep 2026	$550M Series C led by Insight at $5B, Salesforce joins. $170M secondary. 650 staff. Run rate ~$70M.	TechCrunch; CTech
6 Sep 2026	CTech: gross margin ~52%; 400 of 650 staff at customer sites.	CTech
9 Sep 2026	FC Bayern names Wonderful "Official Enterprise AI Partner" for three seasons.	Sportcal; Yahoo Sports

Three rounds in ten months is the number people quote. The number we find more telling is 3.5 months, the time from founding to sales conversations with named enterprises. Everything after it followed from having a customer to show.

Part three: the revenue line, and what it cost to draw

Wonderful does not publish revenue. The Israeli business press does, from what it calls market estimates and sources close to the company, and the sequence is consistent across outlets.

Point in time	Annual recurring revenue	Headcount	Source
August 2025	$1M	~90 (Aug)	Globes, 17 Jun and 12 Aug 2026
November 2025	expected $10M for 2025	120	TechStartups citing Reuters; CTech
End of 2025	$7M		Globes
Q1 2026	$18M	300 to 350 (Mar)	Globes; TechCrunch; PR Newswire
June 2026		~530, 240 in Israel	Globes
August 2026	$55M	~630	Globes
2 September 2026	~$70M run rate, $100M expected by year end	650, about half in Israel	CTech

Annual recurring revenue is the yearly value of the subscriptions a company has signed, and a run rate is the current month multiplied by twelve. Neither is audited here. What the two columns show together is the shape of the business: revenue grew 70 times in a year and headcount grew seven times, which sounds like software until you read where the people sit.

CTech's Sophie Shulman reported on 6 September that about 400 of the 650 work at customer sites, and that the gross margin, citing the Wall Street Journal, is about 52 percent, against the 70 to 90 percent a software company usually keeps. Gross margin is what remains of each dollar after the direct cost of delivering the work, so the missing 18 to 38 cents is mostly the salaries of the people in the customers' buildings.

Two analysts divided revenue by people and got different numbers, because they used different revenue bases. VC Cafe's Eze Vidra put it at about $108,000 per employee on the $70 million run rate. A 3V analysis put it at $154,000 on the $100 million year-end guidance, and noted that at the planned 900 people it would fall to around $111,000. Both are arithmetic on reported figures. Both are far below what a software company of this valuation produces per head, and both writers say so.

Part four: the product, and its three names

The product changed name three times in 14 months, and the changes track what customers turned out to want.

Name one: AI customer-facing agents for non-English markets. The seed boilerplate: "Wonderful builds AI customer-facing agents for enterprises in underserved markets. Wonderful's agents are deeply localized and deliver instant, expert customer support across chat, voice, and email." Voice was the first channel and Hebrew the first language, with Greek, Italian and French next, per CTech.

Name two: the enterprise agent platform, then AI transformation. Winkler told Globes what forced the change: "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 between systems. Page titles on the site moved from "The Enterprise Agent Platform" to "AI Transformation for the Bold Enterprise."

Name three: the Enterprise AI OS. From the Series C onward, every page title reads "Wonderful | The Enterprise AI OS." SiliconANGLE's Maria Deutscher described the pieces on the day of the round: an agent development environment with version control and A/B testing, and a gateway that routes requests to the most suitable language model, with support for Google's protocol for splitting tasks across agents. The company's own module strip on its homepage lists "Data, Alerts, Permissions, Skills, Orchestrator, CLI, Evals, Guardrails, Workspaces, Knowledge." Lalazar's line in the announcement: "We designed the AI OS to be modular and open because enterprises shouldn't have to replace everything they already have to become AI-native."

Three product moves along the way matter for anyone who builds this kind of thing. The Agent Builder, announced on 23 January 2026, is an agent that builds and tests other agents, built on Anthropic's Claude, which the company says "iteratively builds and evaluates agents until they meet production requirements" and cut build times by up to 50 percent across more than 60 deployments, by its own account. Computer use, announced 21 June 2026, lets an agent operate a legacy system through its screen on a dedicated virtual machine where no programming interface exists. And the deployment page lists four ways to run the thing, from shared cloud to "Completely air-gapped, zero external dependencies, with full on-site maintenance by our engineers," which is the sentence a bank's security team reads first.

The most unusual product statement is Lalazar's essay of 10 June 2026, Open by default, competitive by design. In it he writes that "leaving Wonderful should be easy," that every agent, skill, tool and configuration can be exported, that the company publishes an interface file with hundreds of endpoints and a connector that converts Wonderful agents into Azure agents, and, as a consequence, "We cannot raise prices arbitrarily." His closing line: "If we can't win on merit, we shouldn't win." Whether that survives an investor's margin expectations is one of the open questions at the end of this piece. As a written commitment from a chief technology officer, it is rare.

Part five: how it sells

Wonderful's sales model is the part investors paid for, so it deserves its own section.

The pod. The careers page lists three roles and describes each. The Deployment Strategist: "Problem-solving leaders with a can-do mindset and strong technical fluency," who "own deployment and are measured on customer P&L outcomes." The Forward Deployed Engineer: "generalists across integrations, agents, and data, or specialists such as telephony solution architects." The Go To Market seller: "Consultative enterprise sellers" who "coordinate a Wonderful pod." The tagline is "Global by Design. Local by Execution."

The company's own essay on the engineer role, published 3 March 2026, says the closest analogy for the job is a founder: "The FDE is a founder building something new inside a large organization, figuring out what works in the field, creating without a playbook", and describes them as "senior technical owners, embedded in customer environments, who own the technical outcome end-to-end." Omer Perez, its general manager for Israel, described the mix on an Israeli tech show as 60 percent programmer, 20 percent product manager and 20 percent business, according to a write-up we could only read in summary.

The country launch. Each market gets the same treatment: a blog post titled "Wonderful ", a named general manager, and local hiring of strategists, engineers and account managers. The Greece page, the first, says "Wonderful is opening its first international office in Greece, expanding beyond Israel for the first time," and its general manager Ilia Tzortzopoulou calls Greece "a linguistically complex and historically underserved language, where customer support remains costly and limited." The launch posts run from Greece in September 2025 through the Netherlands, Switzerland, France, Italy, Latin America, Asia-Pacific, Singapore, Australia and India in August 2026. The careers page now lists 32 cities, from Tel Aviv to Johannesburg, and the about page names 29 country leads by name.

The Balkans as a test bed. The Recursive ran a February 2026 interview under the headline "Why Is a $700M Startup 'Testing' Its AI in the Balkans?", with Winkler and the Adriatic general manager Vedran Bajer. It is behind a paywall and we could not read it, so we note only the headline and the fact that the company's Zagreb hub, per its Adriatics page, serves markets whose languages mainstream models handle badly. That is the whole thesis in one office.

The field dispatches. The company publishes a series called In Production, described as "a captain's log told from inside the room," written by the engineers and strategists at customer sites. One dispatch describes a national airline where work was progressing slowly until a new chief executive arrived and said AI was a strategic priority, after which, in the author's words, "Blockers that had been sitting for two weeks were cleared" by the next day. Whatever you think of the marketing, it is the only vendor content we have seen that describes the customer's internal politics as the actual bottleneck.

The price. Nobody has published one. No page on the site shows pricing, and no article we found reports a per-seat, per-conversation or per-outcome figure. The company's AWS Marketplace listing shows contract pricing by private offer, with one listed 12-month enterprise line at $2,500,000. CTech reported that the company grows partly because it prices low compared with Salesforce and Accenture. Lalazar's statement that the company "cannot raise prices arbitrarily" is the only public commitment on the subject.

The partners. McKinsey and its analytics arm QuantumBlack announced an alliance in April 2026. Anthropic's Claude powers the Agent Builder, with an Anthropic executive quoted in the launch. The product is listed on AWS Marketplace, and an AWS country lead is quoted in the PPC Energie case study. Salesforce invested in the Series C. FC Bayern signed a three-season partnership in September 2026, reported by Sportcal. We found no reseller or systems-integrator channel of the kind older enterprise software companies use.

Part six: the customers, and which claims stand on their own

The table separates what an independent outlet reported from what the company says about itself. The results column is the company's own in every row, because no customer result has been audited by anyone outside the relationship.

Customer	Where	Relationship reported by	Result, by Wonderful's own account
Bezeq	Israel, telecom	CTech, July 2025; Globes	"Three out of four cases resolved on first attempt"; 6,500 interactions in six weeks
Maccabi Health Services	Israel, health	CTech, July 2025	Agent handling 45,000 calls a day; 1,400 appointments booked a day
Bank Hapoalim	Israel, bank	Globes; IVP	First voice agent live in 72 hours; 75 percent resolution; a bank engineer built the second agent in three weeks
Bank Leumi, Discount Bank, Israel Electric, Pazgas, Menorah Mivtachim, Libra	Israel	Globes; ynet; IVP	No figures published
OTE Group	Greece, telecom	Company case study only	Deflection to 50 percent in eight weeks
ELTA Hellenic Post	Greece, postal	Company case study only	86 percent resolution; calls handled per day from about 500 to about 2,000; five weeks to production
Banco Caja Social	Colombia, bank	Company case study; VC Cafe repeats it	Collections agent live in 19 days; promise-to-pay rate from 45 to 65 percent
Telefónica Colombia	Colombia, telecom	Company case study only	77 percent of billing issues resolved
Petrol Ofisi	Turkey, fuel	Company case study only	Internal IT agent live three weeks after signing; handling time from four minutes to under one
PPC Energie	Romania, energy	Company case study only	Handling time from six minutes to 90 seconds; four weeks to production
Mercado Libre Vehículos	Mexico, marketplace	Company case study only	Three WhatsApp agents in ten weeks; about 99 percent handled without a person
FC Bayern	Germany, sport	Sportcal; Yahoo Sports	Matchday ticket support agent; terms undisclosed

Two things stand out. First, every Israeli relationship is independently reported and every non-Israeli one, except FC Bayern, exists only on Wonderful's site. The 3V analysis made the same point: the disclosed clients cluster in Israeli banking and healthcare, while the "world's largest enterprises" language in the press releases carries no names. Second, the shape of every case study is identical: a clock, then a rate. Bank Hapoalim in 72 hours, Banco Caja Social in 19 days, ELTA in five weeks. We think that is the right shape, and we have said so at length in Ask an AI vendor for its time to production, but it is a shape the company chose, and the numbers inside it are the company's.

Customer counts follow the same pattern. Eight enterprise clients at the seed, per SiliconANGLE quoting the company; "15 market-leading enterprises," per Index's post the same week; "more than 60 enterprise deployments" in January 2026; about 80 large enterprises in June 2026 per Globes; "over 100 enterprise customers" in the company's India post in August. Two counts four days apart disagree, which tells you these are not the same kind of number.

Part seven: what the investors say

Investors write the most about Wonderful, and they are the least neutral, so their words are worth reading for what they chose to emphasise.

Index Ventures, which led the seed and Series A, framed the bet around geography. Hannah Seal, in the Series A post: "Wonderful has moved from concept to global scale in less than a year, which is extraordinary by any measure." The founders, she wrote, had shown "hiring and scaling faster than any of their peers," and each new market adds "linguistic and cultural training competitors cannot replicate." Index's seed post said the non-English markets "collectively dwarf the Anglophone world."

IVP, which co-led the Series A, described the mechanism: Wonderful "builds country-specific teams with local GMs, engineers, and enterprise sales leaders", and uses "forward deployed engineers working directly inside customer environments." It also made the claim that within a year the company was "serving nearly all major banks, insurers, and telcos" in Israel.

Insight Partners, which led the Series B and then the Series C, gave the thesis its shortest form, that "The next wave of AI adoption will come from local depth" before it comes from global scale, and described "full-stack country-specific teams, from local CTOs all the way to agentic engineers that are embedded within customer environments." Insight's co-founder Jeff Horing, in the Series C announcement, said the company "is building an operating system that enables organizations to scale AI across the enterprise."

Salesforce joined the Series C as a new investor. TechCrunch noted it was Salesforce's first investment in the company. VC Cafe called the stake "both validation and a source of tension," since Salesforce sells its own agent platform and its own system of record, and an independent operating layer sits between the two.

The argument the investors are making, in the end, is the one Andreessen Horowitz's Joe Schmidt published in June 2025 as Trading Margin for Moat: that ServiceNow went public at a 63.2 percent gross margin and Workday at 54.1 percent, that both reached the high seventies later, and that the implementation work that cost the margin early is what made the software impossible to remove. All of Wonderful's investors are betting on that essay, whether or not they cite it.

Part eight: what the press says

The funding coverage in TechCrunch is straight reporting: the amounts, the investors, the headcount, the countries, the founder quote. Three outlets attribute the $700 million Series A valuation, which the company never stated, to Reuters, and every later story took it from there. Bloomberg covered the Series B and Series C; both stories are behind a paywall and we did not read them.

The Israeli business press has done the real work, because it has the sources. Globes ran a long profile on 17 June 2026 under the headline "Wonderful: Holding the AI customer's hand", which is where the revenue sequence, the office count and Winkler's most quoted lines come from: "If you want to work with the largest organizations in the world, you need a presence on the ground wherever you operate," and "Everybody wants AI, but you can't just press a button and have it work." Globes also broke the Series C a month early, on 12 August, reporting that the company was seeking $500 million at $5 billion.

CTech's Sophie Shulman wrote the most useful single article, on 6 September 2026, under the headline "Wonderful is growing like Wiz, but its business is very different". It carries the 52 percent margin, the 400 people at customer sites, the low prices against Salesforce and Accenture, the Palantir comparison, and the Uber parallel, "when venture capital helped subsidize prices and accelerate adoption before the company had to confront the economics of operating at scale." Its question is the one every later piece repeats: whether the growth "can eventually translate into the economics of a software company, or whether its AI ambitions will continue to require the cost structure of a services business."

Shulman's ynet version of the story adds a detail that says more about the model than any chart: Israeli consumers "may have interacted with Wonderful's technology without knowing it" at Maccabi, Bank Leumi, Bezeq and the electricity company, and quotes the Israeli AI community figure Uri Eliabayev: "What was truly brilliant was that Wonderful's people quickly understood that once they got a foot in the door of an organization, it needed more and more assistance and guidance in adopting AI." That sentence is a compliment and a warning in the same breath, and the piece leaves it as both.

The rest of the coverage is local. Sixteen of the 27 items on the company's own press page are in languages other than English: Greek, Polish, Romanian, Italian, Dutch, Serbian, Portuguese, Spanish, German. That is the country-team model showing up in the clippings. Brazil Journal's May 2026 piece quotes Winkler on the difference between European buyers, whose first question is regulation, and Brazilian ones, who ask first about business impact and customer experience, and reports a plan to hire 50 engineers in São Paulo.

Part nine: what the critics and analysts say

The criticism we found is about money, and it comes in four forms.

The margin. Forkast, on the day of the Series C, wrote that the forward deployed model "requires substantial headcount investment and unproven scalability at enterprise scale." CTech and ynet, as above, put the number on it.

The repeatability test. Eze Vidra's VC Cafe piece of 3 September is the sharpest. He accepts the model's logic, noting that Salesforce, ServiceNow and Workday "all required significant implementation work before developing scalable platforms," and then sets three tests. First: "fieldwork only becomes a moat when it becomes product. If every engagement starts from zero, the startup is building a consultancy." The metric is "whether the time, labour and cost required for each subsequent deployment decline." Second: as protocols for connecting agents standardise, "orchestration is important without being inherently defensible," so the value has to sit in context, permissions, evaluation and accountability. Third: whether an independent company can own the control plane before the model providers, the systems of record and the consultancies absorb it. His closing line: "Deployment has given Wonderful a credible head start. The question is whether it can turn that head start into infrastructure."

The undisclosed numbers. The 3V analysis lists what the Series C release leaves out: "The release does not disclose gross margin, average contract value, or the share of revenue that recurs versus the share that arrives as deployment work." It also tracks the vocabulary: "enterprise AI agent platform" and "AI customer support company" in March, "AI operating system for the enterprise" in September, and notes that the new category carries a higher valuation multiple. Its recommended metric is revenue divided by headcount, watched over time.

The review gap. An April 2026 review on aicxstack, which says plainly it did not test the product, flagged the absence of meaningful review volume on G2, Capterra or Trustpilot, a narrower integration library than incumbents, and the professional-services requirement for setup. Its price estimate is the reviewer's guess at a category, stated as such, and we do not repeat it.

One more observation belongs to the critics. CTech noted that a $170 million secondary sale by employees and early angels, "less than two years post-founding is unusually early." Whether that is a sign of confidence or of people taking money off the table depends on who you ask, and CTech did not say.

Part ten: what the people inside say

The company's employees have published more than most.

The chief architect, Daniel Sikorskiy, wrote on 10 March 2026 that the company had banned hand-written code four months earlier. "Four months ago, we banned manual coding at Wonderful. Not 'encouraged AI adoption.' Banned." The post describes building "infrastructure that required models to test their own work before a task was considered complete," having a model keep "a notebook inside the codebase: a running log of learned patterns, past failures, and decisions made," and using different models for different work, one for real-time voice pipelines and infrastructure and another for interface work. It claims the Agent Builder was "roughly 90,000 lines of code, built in about two weeks." Nobody outside the company has checked that figure. We read the post closely in A $5 billion AI company banned manual coding, because we work the same way, and the useful part is the order of events: the checks were built before the speed showed up.

Winkler's own essays give the company's view of why the work is hard. In January 2026: "Across nearly 50 deployments in 18 countries, I've seen the same pattern," which is that companies "have an implementation problem," and he calls the whole thing "an operating-model problem." In July: "The real prize was never a lower AI bill. It's an organization operating closer to its full potential." Lalazar's essays argue for openness, fast iteration and buying the infrastructure while building the agents. All four are on the company's blog and all four say the same thing from different angles: delivery is the product.

Employee reviews are thin. Glassdoor listed six reviews for the company on 16 September 2026. From what its listing shows, the praise is about pay and about being close to the biggest enterprises early, and the complaints are about onboarding and the absence of any boundary between work and the rest of life. Six anonymous reviews of a 650-person company are not a measurement of anything, and we could not read the full page, so we report the count and the themes and no more. The job advertisements are more concrete: a Stockholm posting for a forward deployed engineer asked for six or more years of experience, a "high-agency mindset," direct customer work and native Swedish. That is a description of who the company thinks does this job.

The culture statement on the about page says what the company wants to be: "we move fast, take ownership, and solve problems before they grow," "we value independence and resourcefulness, giving people the freedom to make decisions and drive progress without waiting for permission," and "We play to win." The careers page says who it hires: "high-ownership, mission-driven builders and operators who care more about shipping impact in production than theory, titles, or comfort, and who thrive working on-site with customers."

Part eleven: the loose ends

A company that grows from 90 people to 650 in a year leaves seams showing, and we found several. None changes the story. All of them are worth knowing before you repeat a figure.

The Series C valuation is $5 billion in the company's release, in TechCrunch and in CTech. Globes headlined it at $5.5 billion.
Headcount at the Series B was 300 in TechCrunch and 350 in the company's own press release, on the same day.
On the day of the Series A, Winkler told CTech the company had launched in 10 countries. Index's post the same day said "25+" in one place and 30 in another.
Customer count at the seed was 15 in Index's post and eight in a trade site's report four days later.
The company's Germany page names Yves Brunschwiler as general manager. Its about page names Philipp Putz for Germany and Austria.
The Italy page carries a quote attributed only to "General Manager," with no name.
The deployment page's animated counters render as "0+" and an unlabelled "75%" in the page's server HTML, so a crawler or a reader with scripts off sees a zero.
The logo strip headed "Trusted by leading enterprises across industries" uses images with no alternative text, so the customer names in it are invisible to crawlers and screen readers.
Two blog links from live pages returned 404 when we checked, including the McKinsey partnership announcement.
The "95 percent of the world" line that some commentary attributes to the company does not appear anywhere we could find in its own material. The "AI deserts" phrase is Index's.

We list these because our own site is checked against a specification before every change, and a marketing page is a change. A company whose engineering banned manual coding and whose product line includes an evals module might apply the same discipline to its about page. That is a small point. It is also the kind of small point that a $5 billion valuation invites.

Part twelve: the open questions

Five questions decide whether this story ends as a software company or a large consultancy, and none of them has a public answer yet.

Does the second deployment cost less than the first? This is Vidra's test and 3V's metric. The company's Agent Builder claims are its answer, and they are unaudited. The number to watch is revenue per employee as the customer count grows.
What happens to the price when the subsidy ends? CTech says the company prices low against Salesforce and Accenture and has raised more than $800 million to be able to. Lalazar has written that leaving is easy and that prices cannot be raised arbitrarily. A customer should get that written into the contract.
Who owns the control plane? Salesforce, ServiceNow and Microsoft own the systems of record. OpenAI, Anthropic and Google own the models. McKinsey, Accenture and Deloitte own the boardroom. Wonderful is partnered with one from each group and competes with all of them. VC Cafe's point is that orchestration alone is hard to defend once the connecting protocols are standard.
Does the model survive a downturn? Four hundred people at customer sites is a cost that does not fall when a customer pauses. The company's India plan, 1,000 engineers in Mumbai per ynet, is a bet that the model scales with people before it scales without them.
Can the handover be proved? The Bank Hapoalim story, in which the bank's own engineer shipped the second agent, is the strongest claim on the company's site, because it means the capability moved. It is also the one claim a customer could confirm on its own. Nobody has asked one to.

What we take from it

We have written four separate pieces on what Wonderful's story means for a company like ours, and we will not repeat them here beyond the shortest form. Buyers paid for production and never for a model. The proof that opened the money was a named customer, won in the first four months. Doing the last mile with people has a visible price, and the only way that price falls is if each deployment leaves the next one shorter. Our own bet is that AI writing the code, checked by an eval suite written from the specification, is how that happens without hiring 540 people in a year, and we state that as the mechanism we work by, because we have not measured it yet either.

Our thanks to Sophie Shulman at CTech and ynet, the Globes reporters, Rebecca Bellan and Ram Iyer at TechCrunch, Eze Vidra at VC Cafe, and the writer of the 3V analysis, whose work this piece rests on, and to Wonderful's founders and engineers for publishing enough of their own thinking to be read this closely. A company this young has only a record, and the record is public.

Sources

Independent reporting:

CTech, 2 July 2025: seed round, founders' backgrounds, the 50-person Hebrew team, first customers, 3.5 months to sales.
CTech, 11 November 2025: Series A, 120 employees, "10 countries."
TechCrunch, 11 November 2025: Series A.
TechStartups, 11 November 2025: the $700 million valuation and $10 million 2025 revenue expectation, both attributed to Reuters.
Silicon Canals, 11 November 2025: Amsterdam headquarters.
ISBtech, 12 November 2025: Poland launch.
TechCrunch, 12 March 2026: Series B, 300 employees.
CTech, 12 March 2026: Series B, 350 employees.
Brazil Journal, 29 May 2026: Brazil plans and Winkler on European versus Brazilian buyers.
Globes, 17 June 2026: revenue sequence, 530 staff, 28 offices, Winkler quotes.
Globes, 12 August 2026: the $500 million raise sought, 630 staff up from 90, $55 million ARR.
TechCrunch, 2 September 2026: Series C, Salesforce's first investment.
CTech, 2 September 2026: $70 million run rate, 650 staff, more than $800 million raised.
CTech, 2 September 2026, on the secondary: the $170 million secondary.
Globes, 2 September 2026: the $5.5 billion headline, named Israeli customers.
SiliconANGLE, Maria Deutscher, 2 September 2026: product description.
Forkast via Yahoo Finance, 2 September 2026: "unproven scalability."
CTech, Sophie Shulman, 6 September 2026: 52 percent margin, 400 of 650 at customer sites, pricing, Palantir and Uber parallels.
ynet, Sophie Shulman, September 2026: the Eliabayev quote, the Mumbai plan, consumers interacting unknowingly.
Sportcal, 10 September 2026: FC Bayern partnership.
The Hub, Stockholm job posting: the forward deployed engineer requirements.
The Recursive, 10 February 2026: headline only, paywalled.

Analysis:

VC Cafe, Eze Vidra, 3 September 2026: the three tests, revenue per employee.
3V, 3 September 2026: undisclosed metrics, vocabulary shift, revenue per employee.
Andreessen Horowitz, Joe Schmidt, 4 June 2025: Trading Margin for Moat.
aicxstack, 8 April 2026: the untested review noting the absence of public reviews.

Company and investor statements, all self-declared:

Index Ventures, 2 July 2025 and 11 November 2025.
IVP, 11 November 2025.
Insight Partners, 11 November 2025.
PR Newswire, 23 January 2026: Agent Builder.
PR Newswire, 12 March 2026: Series B release.
Wonderful, Series C announcement.
Wonderful's own pages: homepage, about, careers, deployment, press, Greece, Adriatics, and the case studies for Bank Hapoalim, Banco Caja Social, ELTA, Maccabi, OTE, Petrol Ofisi, Telefónica, PPC Energie and Mercado Libre.
Wonderful essays: The Execution Gap No Model Can Close, Unlocking Capacity, Open by default, FDE: The Engineers Owning The Last Mile, Going Codeless, When the CEO means it, Computer use, Wonderful India.
AWS Marketplace listing: the private-offer pricing line.