Avishai Abrahami | 20VC Summary

1. Key Themes

The SaaSpocalypse Is Mispriced for Many SaaS Companies

The public market is applying an undifferentiated discount to all SaaS companies based on AI disruption fear, but Avishai argues the reality is more nuanced. Some companies have genuine moats (trust, data custody) that vibe coding simply cannot replicate, while others are genuinely vulnerable.

"I think a lot of it is because the market don't know how to calculate the danger for SaaS companies with AI. That's the reason. In some cases, I agree. In some cases, I disagree." 00:07:00

Trust as an Unassailable SaaS Moat

The most underappreciated moat in enterprise SaaS is not the product itself but the institutional trust that allows massive organizations to hand over their most sensitive data. This is the real value of Salesforce, and it cannot be vibe-coded away.

"What other platform will JP Morgan trust for their customers' data outside of the JP Morgan closed garden, DMZ secure environment? None. The answer is none. They don't trust anybody else with that, but they do trust Salesforce. And that's huge. That's not a small thing. It's huge." 00:09:48

Vibe Coding Has Real Limits for Complex Business Stacks

Even professional developers inside Wix — the team that wrote the original product — could not fully replicate the business logic of a vertical SMB offering in Base44 after two weeks of trying. This is a concrete stress test of the limits of current vibe coding tools.

"We had a team, they tried professional developers, they tried to do it for a week... we said, okay, we're going to bring a stronger team. Actually, the team that wrote it in Wix. And two weeks later, they still didn't achieve that." 00:11:25

Fine-Tuned Proprietary Models Will Win on Cost and Quality for Vertical Applications

Wix has built its own fine-tuned model stack for both website generation and Base44, achieving performance equivalent to top-tier frontier models at dramatically lower cost. The insight is that domain-specific training data from real user behavior creates a compounding advantage that generic frontier models cannot match.

"We have a huge amount of training data that is happening for us that we're getting on Base44... we think we can actually create a model that is better for Base44 than just the frontier models that are way more generic... we just released our own AI model... we're actually achieving equivalent to the top tier models with that. So the cost of that is dramatically less." 00:15:24

SMBs Are the Underserved AI Beneficiary

Unlike white-collar knowledge workers who fear AI, SMBs are actually eager for AI to help them run their businesses better. Avishai sees a near-term (within one year) wave where AI tools dramatically improve SMB efficiency in scheduling, marketing, and customer outreach — not replacement, but augmentation.

"If you are in a gym, or your personal trainer, your ability to utilize AI to manage your business better, to sell additional products, and to reach out for new customers will be massively enhanced. And not in 10 years, not in five years, next year." 00:34:09

Off-the-Shelf AI Customer Support Does Not Work at Scale

Despite the hype around AI customer support startups, Wix — a company with 3,500 employees and customer support as its largest department — has repeatedly tried and failed with third-party AI customer support solutions, ultimately building its own.

"We actually try not to build our own. We tried, we tested many different things. It doesn'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." 00:32:09

Buybacks Are Underused by Tech Companies and Are a Legitimate Capital Return Tool

Avishai is a strong philosophical advocate for buybacks as a form of shareholder dividend, arguing that companies are spoiled by the option to simply dilute with stock-based compensation and never balance it with buybacks. Despite the terrible timing of Wix's own buyback, the thesis remains intact.

"I think that buybacks are a fantastic tool for a company and companies should use that. And I think they don't use it enough... it's another way of giving a dividend to all of your stockholders." 00:25:49

AI Is Progressing More Slowly Than Advertised — The AGI Definition Has Been Quietly Moved

Avishai has changed his mind on the speed of AI's displacement of humans. He points to the fact that the definition of AGI has been retroactively narrowed to match what models can actually do, rather than the original ambition of solving fundamental scientific and physics problems.

"A year ago, we would hear Elon Musk speaking about he's waiting for AGI because AGI, you can have conversation with it about how to solve big problems in physics... And now everybody's talking about, well, we kind of achieve AGI, but it still cannot do those big questions. So we kind of change the definition to fit something that doesn't do the critical things." 00:50:42

The Dual-Platform Hedge: Owning Both Ends of the Website Creation Spectrum

Wix's strategic position is not to bet on one future but to own the no-code/SMB end (Wix) and the vibe-coding end (Base44), so that customer migration between modes stays within the portfolio. This is a deliberate hedge against disruption from within.

"They might move from here to here. That's fine... We own Base44." 00:12:48

2. Contrarian Perspectives

LLMs Are Not Good at Reasoning — They Are Data Reframers, Not Thinkers

Against the mainstream narrative that AI is approaching human-level reasoning, Avishai argues that LLMs are fundamentally good at collecting and reframing data but not at actual reasoning. He demonstrates this with a concrete example: Claude generated six safety gates for a protocol, and under interrogation, five of them dissolved.

"I needed to test something and I wanted Claude to write me the safety protocol. I got six gates. And he said, those are the gates you have to make sure to test every one of them. So I started to ask why this one and we started to analyze... From six, it became one. All of the other ones said, oh, you know what? You really caught me here." 00:35:10

Atlassian Is the SaaS Company Most Vulnerable to AI Disruption — Yet It Has Recovered

While most observers worry about SMB-focused SaaS being disrupted, Avishai argues that developer-facing tools like Atlassian are actually the most at risk because their users are the very people most capable of building replacements with vibe coding. He is a shareholder and admits the stock's recovery has proven him wrong.

"I think that the customers are the ones that are more likely to build things with Claude Code because they are developers. So I'm always in a bit of a concern there because I'm like, somebody would replace their management stack of the tickets and workflow... And yet you said, this is the one that recovered. So obviously, I'm missing something." 00:10:28

AI Customer Support Startups Are Selling a Product That Doesn't Work at Enterprise Scale

Despite massive valuations and hype, the AI customer support category has fundamentally failed to deliver for a sophisticated, large-scale customer like Wix. The structural reason: the buyer has better engineers than the vendor.

"They proved something very quickly and hired a bunch of salespeople and trying to sell it. And then they're going to one of those companies you mentioned where they have super smart engineers that are really focused on that. They have more engineers supporting their support, right, than all of the HR on those companies that are trying to sell them, and better engineers a lot of times." 00:32:09

One-Person Company Acquisitions Can Be the Best M&A — And No One Asked About It

Wix bought Base44 when it was literally a one-person company for $80 million. The board, rather than questioning the price-per-person absurdity, focused on business logic, go-to-market strategy, and team building. The acquisition has already justified its price multiple times over.

"We bought a company with one person... Luckily, Maor is a brilliant person, amazing leader. So essentially, I don't know how to do another one of those, all right?" 00:21:34

Wix Stock Is Trading on Anthropic and OpenAI News, Not Its Own Fundamentals

A public company generating $400 million in free cash flow with a market cap of $2.8 billion and a rapidly growing vibe-coding product is being valued almost entirely by the news flow of third-party AI companies it does not control.

"Today we're trading on other companies' news. We're not trading on Wix news. We're trading mostly on what OpenAI or Anthropic are saying or Google is saying. And the reality is that I cannot influence that." 00:00:00

3. Companies Identified

Wix Website creation and SMB SaaS platform. The subject of the entire episode — discussed for its $2.1B revenue, $400M free cash flow, 3,500 employees, deep trust-based moat with small businesses, and dual-platform strategy owning both traditional website building and vibe coding through Base44.

"We make about $400 million a year, okay, in free cash flow. So we make cash and we invest about $200 and something million into Base44." 00:00:00

Base44 Vibe coding / AI app-building platform acquired by Wix as a one-person company for $80M, now at 400 employees and over $150M ARR. Praised for its out-of-the-box ease relative to Claude Code, its proprietary fine-tuned model, and its role as Wix's hedge against AI disruption of its core business.

"Base44 gives you a lot coming out of the box so you can do way more with it." 00:11:54

Salesforce Enterprise CRM. Cited as an example of a SaaS company whose real moat is institutional trust and data custody, not the product itself — making it genuinely hard to disrupt with vibe coding despite surface-level CRM functionality being replicable.

"The fact that you trust SaaS providers with your data. You have huge banks, huge companies that actually allow Salesforce to hold all of the customer data because they trust Salesforce." 00:08:49

Figma Design tool. Named as the competitor Avishai most admires — praised for its product quality, market understanding, and the almost religious loyalty it inspires among designers.

"The one company I really, really appreciate, Figma, I think they're brilliant product, understanding of their market... My designers, they love it. They go crazy about it. They'll do their grocery list on Figma." 00:51:47

Monday.com Work OS platform. Mentioned as a parallel success story, with Avishai noting its CEO Eran Zinman as someone who, unlike Avishai, genuinely struggles to detach from the stock price — and who has great surfer hair.

"Aaron from Monday said he was not able to do that." 00:20:54

KHealth AI-powered healthcare diagnosis company. Invested in by Avishai; cited as proof that AI can outperform doctors in routine diagnosis scenarios.

"I'm invested in a company called KHealth... they've proven that they diagnose better than most of the doctors." 00:40:02

Atlassian Developer workflow and project management tools. Discussed as a potential AI disruption risk because its user base (developers) are the most capable of self-building replacements — yet the stock has recovered, which Avishai admits surprised him. He is a long-term shareholder.

"I love the company. I've been a shareholder for a long time." 00:10:28

Isomorphic Labs / Latent Labs / Chai Discovery Scientific AI labs. Cited as the correct category of AI for complex scientific reasoning — specialized non-LLM architectures doing genuine research, as distinct from general-purpose LLMs which Avishai argues cannot do real science.

"Your Isomorphic Labs and your Latent Labs and your Chai Discovery, who specialize in very, very complex reasoning around scientific..." 00:41:35

DeepMind AI research lab. Cited for the AlphaFold protein folding achievement as an example of genuine AI scientific breakthrough — but explicitly a specialized model, not a general LLM.

"You did the protein folding, right? DeepMind did that. Amazing. An amazing achievement, of course." 00:41:54

Lovable Vibe coding competitor. Mentioned in context of why Base44 was considered superior at acquisition — the board asked specifically why Base44 was better than Lovable and Replit.

"We got a lot of questions in regards to that also about why we think it's better than Replit or Lovable." 00:22:51

Replit Vibe coding competitor. Same context as Lovable — named as a benchmark competitor Base44 was evaluated against. Also noted for pursuing multi-agent, multi-model quality maximization rather than cost optimization.

"You see all these hyped customer support companies... like Replit who do like multi-agent and multi-model modes, they are not cost optimizing at all, but they are delivering a good customer experience." 00:31:54

Robinhood Fintech brokerage. Mentioned illustratively by Harry to make the point that founders and CEOs cannot detach their emotional state from stock price — Vlad Tenev visibly transformed between a $20B and $120B valuation.

"I remember interviewing Vlad from Robinhood, okay, and I interviewed him when it was worth like 20 billion and interviewed him when it was worth 120 billion. I can tell you he was legitimately six times happier." 00:50:01

Lagora AI legal services company. Mentioned as a 20VC portfolio investment, used as a discussion prompt around whether AI progress in legal will mirror the rapid gains seen in coding.

"We're an investor in Lagora, which is an AI legal service." 00:38:20

4. People Identified

Maor Shlomo Founder of Base44 (the one-person company acquired by Wix for $80M). Described as a brilliant engineer and strong leader who was capable of anchoring an entire 400-person team being built around him post-acquisition. The single most important hire/acquisition in Wix's AI strategy.

"Luckily, Maor is a brilliant person, amazing leader. So essentially, I don't know how to do another one of those." 00:21:34

Eran Zinman Co-CEO of Monday.com. Mentioned as a peer public company CEO who, unlike Avishai, genuinely cannot detach his emotional state from the stock price — used as a contrast to illustrate Avishai's unusual equanimity.

"Aaron from Monday said he was not able to do that. And so... More white hair, you see that?" 00:20:54

Alon Carmel CEO of KHealth. Mentioned by name as the leader of an AI health diagnostics company that Avishai has invested in, which has proven AI can outperform doctors in routine diagnosis.

"I'm invested in a company called KHealth. I don't know if you're familiar with it. Yeah, yeah, yeah. Alon. Yeah, Alon." 00:40:02

Sam Altman CEO of OpenAI. Cited in the context of the advice to build for tomorrow's model, and as a comparison point for how visionary storytelling differs from Avishai's more grounded Israeli directness.

"I had Sam Altman on the show and he said, you have to build for the model of tomorrow or next year." 00:12:48

Andrej Karpathy AI researcher and former Tesla/OpenAI figure. Cited by Harry as evidence of rapid AI adoption in coding — going from 20% to 80% of coding work assisted by AI in six months — as a benchmark for whether similar jumps will happen in adjacent fields like law.

"Andrej Karpathy went from doing 20% of his coding work with coding tools to 80% in six months." 00:38:20

Vlad Tenev CEO of Robinhood. Mentioned as a vivid example of a CEO whose persona and visible happiness was completely correlated with stock price — the antithesis of Avishai's equanimity.

"I can tell you he was legitimately six times happier when it was worth 120 billion compared to when it was 20. It was a different fucking person." 00:50:01

Chamath Palihapitiya Investor. Cited by Harry as having claimed open-source models are 14-16x cheaper than frontier models for certain use cases — used to challenge Avishai on the cost savings potential of proprietary models.

"Chamath came out and said that, hey, when he uses open source, it's 14 to 16x cheaper." 00:17:09

5. Operating Insights

The Late-Night CEO Schedule as an Organizational Forcing Function

Avishai's 5–6am bedtime, 11:30am wake-up schedule is not just a personal quirk — it has a deliberate organizational benefit. When the CEO is unreachable in the morning, teams are forced to resolve their own disputes, which systematically raises organizational maturity and filters only the genuinely hard problems upward.

"When you're a small company, people want the CEO to solve their problems. By the time I would arrive, most of the things would be solved by themselves. So the thing that were left for me to handle would be the more, well, the harder things to solve... So I think that actually helps the organization be more mature." 00:05:11

Talent Turnover in a Down Market Is a Refresh Opportunity, Not Just a Loss

Rather than treating talent loss during a stock slump as purely negative, Avishai reframes it: forced turnover reveals hidden talent that was previously invisible, and brings in new people without the institutional staleness of 15-year tenures. The real question is not how to prevent departures but how to ensure the people who stay are genuinely excellent.

"It's not about how do you prevent people from leaving. It's more a thing about how you make sure the organization keeps being an organization where the vast majority are top talents... you're going to find you have a lot of talent that you never noticed before." 00:29:27

Context-Switch Discipline: Time-Boxing Work Thinking to Protect Personal Life

Avishai solved the common founder problem of being mentally absent at home by developing a deliberate practice of stopping his work thinking at a set time and scheduling it to resume later — sometimes literally writing a note to think about something in two hours. This is what enables the late-night deep work block and genuine presence with family.

"I stop most of it now and then I'll do it later. The thinking... I'm just doing that and then I'm turning it back on. And sometimes I write myself things that I need to think about in two hours." 00:48:31

Don't Let the Continuous Feedback Loop Die — Retrain Weekly

On Wix's proprietary model for website generation, the competitive advantage is not just the initial training but the continuous weekly retraining loop driven by real customer signals. This compounds over time and is what makes the model faster, cheaper, and more accurate than even the frontier alternatives for that specific task.

"We keep improving it because we have continuous... A continuous improvement loop, feedback loop, because we see when customers like something. And so every week we are trying to run it again and train it again to get to a place where it's better." 00:16:23

6. Overlooked Insights

Wix Is About to Launch an Entirely New Product Next Month

In a single throwaway sentence, Avishai mentions that Wix is about to launch a new product the following month — distinct from both Wix and Base44 — and then moves on without elaborating. This received zero follow-up in the conversation but is potentially highly significant for understanding where Wix's product strategy is heading beyond the dual-platform model.

"We are about to launch another product next month, all right? So that's a different product from... that we're built." 00:21:34

The Real Ceiling for AI Customer Support Is Engineer Quality Asymmetry — Not Technology

The conventional narrative is that AI customer support startups will eventually mature. Avishai identifies a structural reason they may never win at sophisticated enterprise buyers that is almost entirely overlooked: the buyer's engineering team is already better than the vendor's. This is not a temporary gap — it means the vendor's product will always lag behind what the buyer could build themselves, creating a self-defeating sales dynamic for the entire category at the top of market.

"They're going to one of those companies you mentioned where they have super smart engineers that are really focused on that. They have more engineers supporting their support, right, than all of the HR on those companies that are trying to sell them, and better engineers a lot of times." 00:32:09