Wix Founder: What Wall St Gets Wrong About AI & Wix | Will Base44 Win the Vibe-Coding Wars?

Why does it matter? Because Wall Street has priced Wix's core business below zero — and the man running it has the receipts

Wix generates $400M in annual free cash flow and owns Base44, one of the fastest-growing vibe-coding platforms alive. At peer category multiples, Base44 alone should be worth roughly $8B. The entire company trades at $2.8B — which implies the core business is worth less than nothing. Founder and CEO Avishai Abrahami has a sharp case for why the market is wrong about almost everything driving that discount.

• Base44 at vibe-coding peer multiples ≈ $8B; Wix market cap = $2.8B — the core business has negative implied value • Wix's own engineering team — the people who originally wrote the code — spent two weeks trying to build a hairdresser app in Base44 and couldn't finish it • Fine-tuned models save 1–30% versus frontier on complex tasks, not the 14–16x some investors are underwriting • The CEO who built one of the leading vibe-coding platforms grew more skeptical about AI replacing humans over the last 12 months, not less

Base44 at peer multiples is worth $8B — Wall Street has valued the rest of Wix at negative $5B

Assign Base44 the same multiple as any vibe-coding peer and it prices alone at roughly $8B. Wix's entire company — $2.1B in revenue, $400M in free cash flow, 3,500 people — trades at $2.8B. The implied value of the core business is negative.

Abrahami isn't claiming the market is uniformly wrong. He understands why SaaS multiples collapsed and agrees the AI threat is real for some incumbents. The problem is that Wix is being priced on the wrong inputs. "Today we are trading on others companies news. We're not trading on Wix news. We're trading mostly on what OpenAI or Anthropic are saying or Google is saying."

The structural trap for any public company with a high-growth AI division embedded in a legacy business: markets can't cleanly separate the two. Instead of pricing the AI unit as a call option on a breakout, they discount the whole entity. The $400M in annual cash flow funds the build while the market figures it out.

JP Morgan trusts exactly one platform with customer data outside its own servers — and that trust is worth more than any CRM feature

JP Morgan trusts exactly one platform with its customer data outside its own closed, secure environment. "The answer is none. They don't trust anybody else with that. But they do trust Salesforce." That institutional permission took decades to earn. No vibe-coded alternative inherits it, regardless of how capable the underlying model becomes.

The Salesforce bear case — that agents will commoditize it into a data pipe — misses what the product actually is. Even if the CRM becomes a commodity interface, it doesn't matter: "the CRM itself is a very small part of the value." Data custody is the product. "You're not going to vibe for that because the trust is massive."

The same logic extends differently to Shopify — protected not by trust but by complexity. "You're not going to vibe code Shopify no matter how good you are. The business stack is too hard." Companies genuinely at risk are those whose value is pure functionality, easily replicated. The current AI narrative prices institutional trust and operational complexity as if they simply don't exist.

Wix sent the engineers who wrote its original code to build a hairdresser app in Base44 — they couldn't finish in two weeks

First, a professional developer tried for a week. Didn't get close. Then Wix escalated: they brought in the team that originally wrote the Wix codebase itself. Two weeks later, still not done.

The test case was a hairdresser management system — scheduling, staff, client operations. Not exotic. This is the company that owns one of the leading vibe-coding platforms, purpose-built to make exactly this kind of thing easier than anything else on the market. "I don't think the pizza place next to your house is going to use Claude to build his delivery, his scheduling, his staff management system."

Business logic is an entirely different problem from building a website. Base44 compresses that gap significantly — it ships far more out of the box than any general-purpose AI coding tool — and still lost the test. TAM erosion from vibe-coding adoption is real, but the majority of SMB owners are not about to become citizen developers. Not in the near term. And if they eventually do, Wix owns Base44.

The CEO who built a leading vibe-coding platform spent 12 months becoming more skeptical about AI — not less

"I was very concerned with how quickly we're going to find humans being replaced with AI. And I think I'm now looking at the same thing. I think it's going to take much longer than I anticipated."

This is someone running fine-tuned proprietary models, watching failure modes at production scale across hundreds of millions of sites, with every commercial incentive to want AI to work faster. His specific diagnosis: it's "very good at collecting data and reframing it" but "not very good at reasoning." The AGI definition has been quietly rewritten to match what models can actually do, and he noticed. "We kind of change the definition to fit something that doesn't do the critical things. The more I know about AI, the more I see the frontier model, the more I see things — they're tremendous... but I don't think they're going to replace us that quickly."

The concrete test: he used a frontier model to generate a safety protocol, got six gates, challenged each one, and watched five retracted in real time. That's not reasoning. That's approval-seeking.

Their fine-tuned model matches frontier quality on Base44 tasks — but the savings are 1–30%, not the 14–16x investors want to believe

Between 1% and 30% cheaper. That's what Wix gets running its fine-tuned domain model versus frontier API calls for Base44's core generation tasks — not the 14–16x savings Chimath Palihapitiya cited from open-source switching. Abrahami's rebuttal is direct: those figures come from swapping frontier inference for lightweight models on simple tasks. "When you want to use a complicated model — for example, what you need for something like Base44 — it's not going to get to that much of a level."

The model strategy is still defensible on quality grounds. Frontier models carry knowledge of Chinese poetry that's completely irrelevant to interpreting a clumsy natural-language request for a task manager. A domain model trained on actual user intent, improved through a weekly feedback loop, can outperform frontier at the specific task that matters. They've already demonstrated this on the core Wix product — generated websites now run on their own trained model, faster and with fewer errors. The margin story is real. The magnitude the bulls are underwriting is not.

Wix tried every off-the-shelf AI customer support vendor, built its own, and still can't make it work

"We built our own. It doesn't work."

Three sentences later: "We actually tried not to build our own. We tried many different things. It doesn't work."

Wix operates support across 192 countries. Customer support is their largest department. They have direct financial incentive to solve this and deeper engineering resources than most of the startups currently pitching enterprise AI support solutions. They cannot get it to work.

The failure sequence in his own words: "We tried. It didn't work. We tried again. It didn't work. We tried again. It didn't work. We kept working on our solution on the side. At some point, it was like — okay, that's enough."

His diagnosis of why startups keep failing at the enterprise boundary: they prove a demo fast, scale sales, then land at companies with better engineers who've been working the problem longer and have far harder operational requirements. Maybe five years changes this. Right now, across 192 countries, it doesn't work.

Freedom from fear makes better decisions — the CEO who chose to show up this morning operates on a different quality curve

The real value of money, in Abrahami's telling, isn't lifestyle. It's that "freedom also to know that you are here because you've chosen to be here." He could stop. He doesn't — which means every decision he makes carries the weight of a genuine opt-in rather than a trapped one.

"If I'm here, I need to be fully committed to what I'm doing because I've chosen to be here. Nobody made me." Stock crashing, narrative hostile, entire sector in the wrong column — he wakes up focused on what he can control, because stepping away is a genuine option, not a fantasy.

The downstream effect: a CEO operating from fear of losing their position makes defensive, short-term calls designed to preserve it. One who has genuinely opted back in every morning can absorb multi-year pressure without folding the hand. The market prices this psychological difference at zero. It isn't.

High stock prices were quietly locking mediocrity alongside top talent — attrition is the correction, not the crisis

"A lot of us got very spoiled — we have people here working for 15 years that are always top talent, nobody ever leaves because our stock is whatever." Strong equity prices are an indiscriminate loyalty mechanism. They keep the excellent and the comfortable in the same place, invisibly. When the stock falls and people start leaving, you see the floor for the first time.

Abrahami's counterintuitive advice to fellow CEOs isn't to panic-retain everyone. It's to be deliberate about who you actively keep. "The people that you allow to stay are really great people — you're going to find you have a lot of talent that you never noticed before."

He's not dismissive of real losses — there are people at Wix whose departure would genuinely worry him. But the frame is unambiguous: the question isn't how to stop attrition. It's how to maintain quality through it. Every forced turnover period is an opportunity to reset the talent bar. Most CEOs spend it trying to preserve headcount instead.

The companies that survive the next phase own both the moat and the alternative — and that list is very short

Wix is reporting consistent failure in a category — enterprise AI customer support — that's attracting enormous venture funding. Its fine-tuned models save far less than the gross margin expansion narrative requires. Its own platform couldn't handle a hairdresser's operations after two weeks of expert effort. The CEO most incentivized to believe the AI maximalist case has grown cautious.

The implication cuts across all of it: the companies positioned to win aren't the ones betting that AI obliterates their core product fastest. They're the ones that own both the institutional moat and the vibe-coded alternative. Wix, whatever the market currently thinks, is that company.

The math is either the most obvious mispricing in tech right now — or the market knows something Abrahami doesn't.

---

Topics: SaaS, AI, vibe coding, public markets, enterprise software, Wix, Base44, founder interviews, LLMs, SMB, AI valuation, customer support AI, model fine-tuning, leadership, talent retention