Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
Summary
Abrahami argues Wix’s $2.8 billion valuation assigns Base44 less than its implied standalone value and the core business less than zero. Against roughly $2.1 billion of Wix revenue and Base44 already above $150 million ARR in the host’s framing—the host suggests roughly $160 million, and Abrahami says “more”—he estimates Base44 alone “should probably be at 8 billion.” His wider call is that markets are pricing AI disruption without separating replaceable interfaces from trusted, operationally complex systems.
Wix’s defense against AI cannibalization is to own both the packaged-software and vibe-coding paths. Professional developers spent a week trying to recreate just the business logic for a hairdresser in Base44; a stronger Wix team still had not finished two weeks later. Some customers will migrate, but Abrahami’s answer to the TAM-reduction pushback is blunt: “We own Base44. They might move from here to here. That’s fine.”
Base44 currently has worse margins and retention than Wix, but Abrahami expects inference economics to become a secondary issue within two years or less. Wix website generation and hosting cost “pretty much zero” by comparison, while Base44 consumes expensive models; Wix is countering with specialized, fine-tuned models trained on Base44 usage, failure and intent data. Even so, he would choose “20% more quality” over 20% lower cost because this is still a new market that must first work better.
The $80 million purchase of a one-person Base44 became a heavily integrated Wix build, not a conventional acquisition. Base44 now has roughly 400 people, revenue above the host’s roughly $160 million figure, and could reach 800–1,000 employees within two years, mostly engineers. Wix generates about $400 million of annual free cash flow and is investing “200 and something million” into Base44; the binding M&A constraint is now execution, not financing.
Abrahami concedes Wix’s buyback had “terrible timing,” but judges it on a three-year horizon rather than the subsequent crash. Wix had about $1.5 billion sitting idle, was not planning another major acquisition because of its focus on Base44 and a new product, and bought shares when management considered them cheap. He still sees buybacks as an underused shareholder distribution that can offset stock-based compensation, without minimizing the timing error.
Abrahami has become less worried that AI will replace humans quickly, while remaining bullish on narrowly engineered AI products. His evidence is repeated brittleness: a Claude-generated safety protocol collapsed from six mandatory gates to one after questioning, and systems now described as AGI still cannot answer the hard questions once associated with that goal. Yet he expects AI embedded in purpose-built SMB workflows to improve selling, scheduling and customer outreach “next year,” because “it’s amazing, it’s incredible, it’s super powerful—but the but is pretty big.”
The operating thesis is ultimately execution, not the stock tape or an AGI forecast. Wix trades on statements from OpenAI, Anthropic and Google that Abrahami cannot control, so he says, “I really don’t care” whether shares move 20% either way. His largest concern remains making teams more efficient and ambitious—and ensuring that, having freely chosen the job, he is “fully committed” to it.
Deep dive
1. The market is pricing AI fear before distinguishing which SaaS value can be copied
Harry frames the valuation problem starkly: roughly $2.8 billion of market capitalization against about $2.1 billion of Wix revenue, with Base44 already well above $150 million ARR. Abrahami says comparable vibe-coding multiples imply Base44 alone “should probably be at 8 billion,” meaning the market assigns the established Wix business less than zero.
Abrahami accepts the broad mechanism behind the “SaaSpocalypse”: investors do not know how to calculate AI risk, and some SaaS products genuinely are vulnerable. His disagreement is with applying one answer everywhere—markets cyclically get companies “completely wrong, completely right,” and only time will reveal which fears were justified.
Salesforce is his counterexample because its core asset is not merely CRM functionality but institutional trust. JPMorgan may permit Salesforce to hold sensitive customer data while trusting no interchangeable vibe-coded system; “You’re not gonna vibe code that,” because changing this security relationship is extraordinarily difficult.
Harry’s pushback—worth keeping—is that Salesforce may be a one-of-one trust franchise, while other vendors lack that moat. Abrahami then flags Atlassian, which he owns, as more exposed because developer customers are precisely the people capable of replacing ticketing and workflow tools with Claude Code; its recovery therefore tells him he may be missing something.
2. Wix’s SMB complexity is the moat, while Base44 captures the defectors
Wix primarily serves businesses such as pizza shops and hairdressers whose owners want to run their trade, not assemble delivery, scheduling, staffing and website systems. Abrahami allows that “some, yes” will build these tools themselves, but sees no evidence that the majority will absorb that effort soon.
The sharpest evidence came from Wix’s own experiment: professional developers tried for one week to build only a hairdresser’s business logic in Base44 and achieved little. A stronger team already inside Wix continued for another two weeks without completing it. “Those things are complicated.”
Harry converts the concession into a TAM question: if 20% of customers adopted Lovable, Replit or Base44, would the old market simply become 80% of its former size? Abrahami’s rebuttal is portfolio logic rather than denial: “We own Base44. They might move from here to here. That’s fine.”
His likely five-to-six-year scenario is a spectrum. Low-effort customers remain on packaged Wix products; customers wanting bespoke systems move toward Base44; many sit between them. Wix is already encouraging agencies to use Base44 because visual editing and complex business stacks remain difficult in pure vibe coding: “You’re not gonna vibe code Shopify, no matter how good you are.”
3. Base44 sacrifices current economics for quality and proprietary learning
Abrahami is direct that Wix is substantially more profitable today. Generating and hosting a Wix website costs “pretty much zero” by comparison, and Wix retention is better; Base44 applications demand more work to finish and materially more model expenditure.
Costs are nevertheless declining continuously. Base44 has released an internally tailored system combining several fine-tuned models, which Abrahami says can match top-tier models for its task at dramatically lower cost. Wix already follows the same pattern for website generation, where its own model is faster, cheaper, more accurate and retrained through a weekly feedback loop.
The strategic input is Base44’s usage data: it sees what customers attempt, where generation fails and what users meant despite poor prompts. That permits a narrower model that need not master “Chinese poetry” but can better interpret requests such as building a task manager “to tell my boyfriend where he’s wrong.”
Pressed on savings, Abrahami gives a broad range of 1%–30% cheaper, then says that complicated Base44-like work may yield only 5%–10% savings, not Chamath Palihapitiya’s cited 14–16x. His priority remains unambiguous: offered 20% more quality or 20% lower cost, “I’ll definitely go for the quality.”
4. Base44’s acquisition worked because Wix supplied the company around the product
Harry’s framing is intentionally provocative: Wix paid $80 million for a one-person company. The board discussion was surprisingly easy because directors focused on the business logic—marketing, team construction and why Base44 could beat Replit or Lovable—not the optics of paying so much for one individual.
Abrahami emphasizes that Wix did not acquire a ready-made organization. It bought a product and exceptional leader, then moved substantial Wix talent around it; Base44 now employs about 400 people. That integration success also explains why he cannot casually repeat the deal while preserving the same operational quality.
The revenue payoff has already exceeded Harry’s roughly $160 million figure, with Abrahami responding simply, “More than that.” Yet he resists granular quarterly separation because Wix deliberately moves customers between its platforms, making clean segment numbers difficult.
Wix generates about $400 million of free cash flow annually and invests “200 and something million” or slightly more into Base44. Abrahami says financing is therefore not the M&A constraint: Base44 integration and another internally built product launching the following month make execution the scarce resource.
5. The buyback was mistimed, but Abrahami still defends the instrument
Abrahami’s retrospective begins without euphemism: “Obviously we had terrible timing.” Wix held approximately $1.5 billion, knew it was not going to pursue another major acquisition while focused on Base44 and a new product, and judged leaving the cash idle irrational, so it repurchased shares when management believed the stock was unusually low.
The market initially jumped and then crashed. Rather than claim the outcome was good, he shifts the measurement period: “The big question is not what happened in the last three months; it’s where it’s gonna be in three years.” He hopes—not asserts—that the decision will ultimately prove intelligent.
He remains strongly pro-buyback, describing it as another form of dividend and a mechanism for balancing dilution. Technology companies need stock-based compensation to recruit and retain people who are “invested in the dream,” but cash-generating businesses can use buybacks to balance the shares issued to employees.
His capital-allocation menu is constrained: acquisitions frequently fail and can only be integrated at a limited pace; buybacks return value and offset dilution; hiring more engineers may improve products but destroys EBITDA, which public markets punish. His obligation is to balance shareholders, customers and employees, not react to an imaginary single “stock market” voice.
6. Talent churn can renew Wix, but customer support exposes AI’s limits
Abrahami rejects a magical answer to falling equity values and employee departures: “Everybody will lose talent, and we are losing talent.” The job is not to prevent every exit but to preserve a high concentration of exceptional people and uncover talented successors whom a stable organization may have overlooked.
Long-tenured teams spoiled SaaS leaders into assuming top performers would remain for 15 years. Those veterans originally arrived without 15 years of company-specific knowledge; what mattered was their underlying ability. Abrahami is working intensely to retain critical people while treating the next generation of talent as an organizational necessity.
Wix has about 3,500 people in total, while Base44 has roughly 400, with customer support Wix’s largest department because it operates across 192 countries. Harry suggests AI could reduce support almost to ten people; Abrahami rejects the scale assumption and doubts human support will ever reach zero.
Wix tested multiple support-agent vendors while developing internally. His verdict on the homegrown version is equally candid—“We built our own. It doesn’t work”—but external attempts also failed repeatedly. Young vendors often face customers whose support-engineering teams are larger and stronger than the vendor itself; five years may change that, but Abrahami immediately hedges: “Maybe. I’m not sure.”
7. AI replacement is moving slower than Abrahami feared
Asked what changed in 12 months, Abrahami says he was previously “very concerned” about rapid human replacement and now expects it to take much longer. The AGI goalposts moved: systems once expected to solve major physics problems are now called AGI despite remaining unable to answer those questions.
His capability judgment is deliberately two-sided. LLMs are powerful at gathering and reframing information but weak at reasoning and prone to silly mistakes; another two or three genuine algorithmic jumps might make humans “a bit like monkeys,” yet current systems do not justify assuming that jump has occurred.
His safety-protocol example carries the warning. Claude proposed six mandatory testing gates; after Abrahami interrogated each one, five were demoted or removed, with the model repeatedly admitting it had overstepped. Fluent confidence encouraged “massive trust” that the underlying reasoning did not deserve.
Narrowly engineered systems are different. Abrahami expects AI to help a gym or personal trainer manage schedules, sell additional products and contact prospects as soon as next year. Conversation itself was impossible at today’s quality four or five years ago; connecting it to the right skills and tools can produce a “massive extra layer” without replacing an entire workforce.
8. LLM usefulness depends on whether the domain is fundamentally textual
Law resembles code because statutes, cases and legal work exist in an enormous textual corpus, making it a strong candidate for AI assistance. Abrahami expects substantial usefulness there, while contrasting it with medical research, which he considers a poor fit for a general LLM.
Harry says ChatGPT may already be better than the majority of doctors and notes that health is its second-biggest usage. Abrahami cites portfolio company K Health as having demonstrated better diagnosis than most doctors; routine diagnosis from known scenarios is categorically different, however, from conducting medical research.
Medical research is where he becomes alarmed: an LLM may treat a Reddit assertion and a Nature paper as equivalent material inside one learned representation, without reading or validating the research when answering. Astrophysics and protein folding likewise require simulations or specialized architectures—such as DeepMind’s protein-folding work—not merely a frontier LLM producing plausible text.
9. A public CEO can control execution, not OpenAI’s news cycle
Abrahami says he has detached his happiness from Wix’s stock price: “I really don’t care.” The company now trades largely on news from OpenAI, Anthropic and Google rather than Wix-specific events, none of which he can influence; even during rallies, unresolved operating problems mattered more to him than a 20% move.
His unusual schedule supports that focus: sleep around 5:00–6:00 a.m., wake around 11:00–11:30, and reserve roughly four quiet nighttime hours for real work. Arriving later also forced teams to resolve ordinary disputes themselves, leaving him the genuinely hard problems and, in his view, making the organization more mature.
Resilience begins by accepting that shocks are certain and nonlinear: COVID, war or a market crash can appear on “some random Wednesday.” AI was an exception because the extreme change arrived in slow motion. Preparation means asking honestly whether he is doing the best possible work with the weapons already available.
His remembered Babylon 5 line captures the method: “We’re gonna get there when we get there. And the weapons that we have are the weapons that we have.” Meditation and neuro-linguistic programming also help, but relief primarily comes from separating controllable execution from outcomes no CEO can command.
10. Money’s real return is freedom—and the obligation to choose deliberately
Abrahami agrees that wealth reduces fear and can improve decisions, while adding that it may also make people lazier. Its highest value is freedom: freedom from worrying about food, freedom to leave for Costa Rica or Brazil, and knowledge that “I’m here because I’ve chosen to be here.”
That freedom sharpens rather than weakens commitment. Nobody forces him to endure the stock crash, headlines and operating pressure; choosing them means he must be “fully committed.” His fallback perspective is equally grounding: happiness is not necessarily correlated with wealth, and “at the end of the day, I can always go to Brazil.”
At home, he learned to time-bound unfinished thinking, write down what needs attention and resume it during his late-night hours. His wife modeled the complementary skill by identifying a bad mood as her own rather than loading it onto him: “You did nothing wrong… You go play PlayStation.”
His parenting advice follows the same presence principle: children care less about total hours than the quality of those hours, so arrive happy, engaged and off the phone. Professionally, his enduring concern is also quality of attention—execution, efficiency and ambition—and the kindest gift he remembers from people ahead of him in business was their “most precious currency”: time.