In 8 months, one founder went from bootstrapped idea to an $80 million acquisition by Wix—without a team, without funding, just AI and conviction. In this recent 20VC Podcast Maor Shlomo, founder of Base44, explains why he believes most SaaS companies will be eliminated, why margins don't matter in the AI era, and where smart money should go today.

Maor Shlomo built Base44 as a bootstrap business "just for fun" after his previous capital-heavy startup. What started as a simple tool to help his partner's tattoo business quickly exploded to 100,000 users, then millions. Today, Base44 serves millions of users and is on track to hit $50 million ARR.

His thesis is highly provocative: we're heading toward a world where it's easier to build your own custom Salesforce CRM than to buy an off-the-shelf license. The one-size-fits-all SaaS era is ending, replaced by "vibe coding"—where anyone can prompt an AI to build exactly what they need.

In this wide-ranging conversation, Maor unpacks why he sold despite meteoric growth, how Base44 thinks about defensibility when features copy in weeks, why Google terrifies him more than any startup competitor, and what actually matters when evaluating AI businesses.

For product leaders, founders, and investors navigating the AI platform shift, this is an interesting listen.

20VC with Harry Stebbings

💡 Top tip - This is a dense conversation covering market dynamics, competitive strategy, and investment thesis. Read the executive summary below, then jump into the sections most relevant to your role (defensibility for product leaders, investment criteria for VCs, or market dynamics for founders).

Key insights from the full article:

When Base44 hit strong profitability as a one-person bootstrap, Maor faced three paths: stay solo, raise VC, or partner with a larger company. He chose Wix.

The logic: "This is going to be the largest category in software. Partnering with Wix lets us take a shot at building something huge. Without it, my chances would be really small."

Wix provided marketing muscle, customer support, and operational infrastructure—everything Maor didn't want to build—whilst keeping the product team lean and independent. The company tripled his distribution. When Harry asked if he regrets selling given Base44 now exceeds $100M revenue, Maor said: "Absolutely not."

Key Takeaways:

Maor's provocative belief: most SaaS categories will shrink as it becomes easier to build custom software than buy off-the-shelf.

"There will be a time where it will be easier to build your own Salesforce-type CRM than to buy a license," he argues.

The thesis emerged from frustration. Maor's partner, a tattoo artist, needed a CRM. They tried two SaaS platforms to configure her workflow. "It was hell on earth. Just configuring it was insane." An LLM could build something leaner, simpler, and perfectly tailored—without feature bloat or enterprise settings she'd never use.

Software becomes "liquid." Users start with templates, then vibe code their own features. "You say 'I want this in Arabic, right-to-left, add pictures of each lead.'" The software transforms to exactly what you need. You own the code, own the data, avoid vendor lock-in, and skip features you'll never use.

When Harry pushed back—"Do small businesses care about owning their data?"—Maor clarified: It's not about code ownership philosophy; it's about simplicity and customisation. Small businesses want tools that match their process, not enterprise platforms they contort themselves around.

Key Takeaways:

Features copy in weeks, not years. "Every feature we put out, we know it's going to take a few weeks or months for competitors to copy," Maor admits. Founders from Salesforce, Atlassian, Cloudflare, and Figma have all spun up Base44 equivalents in three months.

Maor's response: "It's relatively easy to create a vibe coding tool. It's very hard to create a platform that helps people build complex, production-ready applications."

Anyone can generate a Monday.com clone from a prompt. But scaling to millions of lines of code, real databases, scheduled tasks, integrations, and user management? That's the moat.

Base44's defensibility is vertical integration: built-in backend, databases, authentication, analytics, integrations, and compute. "You're almost building a small cloud," Maor explains. Critically, they didn't use third-party providers like Supabase or Neon—they built their own infrastructure. "Very hard to replicate and migrate millions of users from Supabase to home-grown."

Velocity still matters for UX and taste—Base44's "app suggestions" feature was copied in days—but foundational infrastructure bets compound and can't be easily replicated.

Key Takeaways:

When Harry asked about competitors—Lovable, Bolt, Replit—Maor surprised: "I'm not worried about those three."

Who worries him? "Model providers. If one wins by a wide margin, the next logical thing is to conquer vibe coding, because this will be the largest software category."

Google is the existential threat. "If Gemini wins the race, they have the entire stack—Google Cloud, data, integrations with Google Suite. They could build an incredible empire."

If you control the best model and the market consolidates, why not move up the value chain? OpenAI has ChatGPT, Anthropic has Claude Code, Google has Gemini across Gmail and workflows.

Maor's hope: continued competition. "As long as there's a fight, we're positioned to control a large share." Multi-model strategies win when no provider dominates. Switching is unprecedented—one line of code moves millions in spend from Claude to GPT-5 to Gemini overnight.

This creates opportunity and fragility. Platforms orchestrating models win. But if one provider storms ahead for six months, revenue collapses for others.

Key Takeaways:

When Harry pressed on margins, Maor deflected but made the strategy clear: "Margins are the least thing I'm worried about."

Base44's strategy assumes model costs trend toward zero. "I could optimise costs now and invest people to do that. Or I'm taking into consideration that prices are just going down and down."

This creates counterintuitive decisions. Base44 deliberately skips some optimisations because future model improvements will solve them for free.

The margin improvement playbook has three layers:

1. Intelligent routing. Route simple prompts ("change button colour to red") to cheap open-source models, not expensive frontier models.

2. Open-source improvements. As smaller models improve, they handle more requests. Margins improve drastically.

3. Train your own model. Cursor released Composer—their own model. The strategy: grow fast with expensive frontier models, gain distribution, then shift to your efficient in-house model. "Overnight, margins get tremendously better."

Base44 is considering similar: "Fine-tune on the billions of lines of code we have, then forward prompts to our fast, high-margin model."

For now, the game is growth and capture, not margin optimisation. "0 to $10M ARR in a year is insane growth."

Key Takeaways:

The LLM market is often compared to cloud providers, but Maor highlighted a critical difference: you cannot switch cloud providers with a line of code. "My previous company tried switching. It was so painful - a year-long process. I probably wouldn't have done it."

For LLMs, it's the opposite. "Claude comes out with Sonnet 4, you're like 'perfect, let's use that.' Then OpenAI comes out with GPT-5, and you move workload."

Base44 uses different models for different tasks. "Claude is the best designer, good at structuring projects. But GPT-5 is really good at solving hard bugs. We move a lot of workload to OpenAI."

This creates unprecedented dynamics. "What happens if one storms ahead for six months? Overnight, Cursors and Cline codes switch to Gemini or GPT-5.5. Such a fast decline—but they also had such a fast ascent."

For model providers, defensibility is existential. Unlike cloud migrations, LLM providers face instant revenue shifts based on performance.

The volatility creates imperatives: Model providers: Move up the stack (ChatGPT, Claude Code, Gemini across Gmail). Platforms: Play multiple providers. Investors: Revenue can swing wildly—$50M ARR today, $20M in six months if a competitor's model leaps ahead.

Key Takeaways:

Traditional SaaS metrics - MAU, churn, NPS - don't fully capture vibe coding success. Maor's surprising metric: sentiment analysis of chat messages.

"We measure sentiment of messages users send, minute by minute," Maor explains. "Many speak with the chat as they would to a human. 'Fantastic job, this is exactly what I wanted,' or 'This is not what I wanted. You deleted this button.'"

Base44 processes tens of thousands of prompts per minute and measures negativity in real time. "When we jumped from Sonnet 4 to Sonnet 4.5, you saw a major improvement."

The metric evolved. Early on: bugs - how often the model broke applications. "Nowadays, it's harder to create bugs. LLMs are smart. Infrastructure is battle-tested. So it's more about: does the agent do what the user asks?"

In early tools, success meant "doesn't break." Now: "understands intent and executes precisely."

Another metric: feature enablement. When Base44 added scheduled tasks, they tracked users who previously hit walls. Adding infrastructure unlocked use cases, measurable in sentiment lift.

Traditional metrics still matter, but sentiment provides real-time intelligence that traditional metrics lag.

Key Takeaways:

Harry's critical question: "Everyone is posting insane revenue numbers. 100 million, 200 million. How much weight should I place on revenue today?"

Revenue growth is exciting—0 to $10M ARR in a year is "insane." But raw numbers miss two questions:

1. Will this get eaten by model providers?2. Is this vertically integrated, or a thin layer on top of someone else's infrastructure?

The "prompt wrapper" problem is real. Jasper AI grew fast but had no moat. "Clearly just using LLMs."

Contrast with vertically integrated platforms. Base44 built entire infrastructure—mini cloud, databases, user management, integrations, compute. "That makes sense to have a business around it."

The Cursor strategy offers a blueprint. Cursor grew fast with frontier models, gained massive distribution, then released Composer—their own efficient model. "Overnight, margins get tremendously better, but you've grown fast and have distribution."

Maor's investment criteria:

✅ Fast revenue growth (0 to $10M+)

✅ Vertically integrated (infrastructure beyond prompting)

✅ Self-sustaining potential (own model, own infrastructure)

❌ Pure prompt engineering

❌ UI/UX layers on commoditised APIs

❌ "Most value is in how they prompt"

Revenue velocity matters, but only paired with structural defensibility. A $50M ARR wrapper on OpenAI's API is riskier than a $10M ARR business with vertical integration.

Key Takeaways:

When Harry asked where smart money should invest, Maor's answer was unorthodox: "The non-sexy industries. Finance, restaurants, HMOs. Entrepreneurs building end-to-end."

AI creates opportunities to build vertically integrated businesses in previously unattractive categories.

"Five years ago, building a law firm would seem like the worst idea. Or buying a hospital and optimising it with AI. Or building a new bank," Maor argues. "Today, you have this huge platform shift. What's the craziest thing we can aim for that's barely possible today but should be possible in two to four years?"

Most software entrepreneurs want to build software, not operate law firms or hospitals. Massive white space for founders willing to integrate AI deeply into non-tech industries.

Where Maor avoids: Fast-growing startups where he can't assess defensibility. Example: AI agent companies. "I don't know if this will get commoditised by model companies." OpenAI, Google, and Anthropic are all building agents. "I'm not sure this category has a lot behind it rather than prompting and tools that will be part of ChatGPT or Gemini."

The pattern: Capabilities that feel like they should be in foundation models (agents, workflows, content generation) are risky. Vertically integrated businesses in complex, regulated, operationally intensive industries create defensible value.

Key Takeaways:

Will we hit a revenue speed bump where AI doesn't deliver the economic value promised?

Maor's position: "I don't believe we'll hit a speed bump. Even with existing models, we're only scratching the surface of economic value you can create."

His evidence is autobiographical. "If a one-person team can get sold for $80 million and build a viable, profitable business, it means you'll be able to do a lot more with less people."

The transformation is structural:

1. Companies become radically more efficient. Maor's previous company raised $130 million, needed 100+ people. Base44 was one person, reached profitability faster.

2. Products get better and cheaper. Efficiency savings flow to consumers.

3. Technical barriers to entrepreneurship collapse. "There's no technical barrier anymore." More startups, more innovation, better outcomes.

The model improvement question is secondary. "I don't think LLMs will hit a wall. But even if they do, we're only scratching the surface."

Value creation is measurable today. Base44 applications range from marketing platforms with AI agents to organisational tools to WhatsApp personal assistants.

For sceptics: "The question is whether better models benefit the entire ecosystem (platform play) or just model companies." The evidence suggests platform dynamics. Cursor, Base44, Notion improve when Sonnet 4.5 or GPT-5 ships.

The macro bet: AI creates compounding value through efficiency, lowered barriers, and better/cheaper products. Even without further improvements, we've barely begun extracting that value.

Key Takeaways:

Maor's journey from one-person bootstrap to $80 million exit in eight months isn't just inspiring—it's a preview of how the software industry is being restructured.

For founders:

Build for your own problems. Maor built Base44 for his partner's CRM needs. "Easiest way to get usable software." Products from personal pain points iterate faster, resonate deeper.

Embrace radical iteration. Throwing away tens of thousands of lines of code is hard but essential. "Once you figure out the product, with a few prompts you get to where you got with hundreds."

Bet on hard-to-copy infrastructure. Features copy in weeks. Vertically integrated platforms with backends, databases, compute take years to replicate.

For investors:

Look beyond revenue to defensibility. 0 to $10M ARR is exciting, but ask: "Will model providers eat this?" and "Is this integrated or a wrapper?"

Favour vertically integrated businesses, even in non-traditional sectors. AI-native law firms, hospitals, banks may offer better risk-adjusted returns.

Margin dynamics are temporary. Model costs trend toward zero. Intelligent routing or in-house models can transform economics overnight (see Cursor).

For product leaders:

Instrument conversational signals. Sentiment analysis predicts PMF faster than MAU or NPS.

Build infrastructure, not features. Features are table stakes. Infrastructure they can't replicate—compute, integrations, databases—becomes your moat.

Play multi-model strategies. Use Claude for design, GPT for debugging, Gemini for cost efficiency. The platform orchestrating models wins.

The meta-game:

The AI shift isn't just creating better tools—it's enabling entirely new business models. One-person $50M ARR companies. Non-tech industries (law, healthcare, finance) becoming venture opportunities. Software becoming "liquid"—customisation replaces configuration.

Winners build the hardest-to-replicate infrastructure, capture distribution, and integrate deepest into industries others ignore.

As Maor says: "If a one-person team can get sold for $80 million, the entire economic value chain is being restructured. We're only scratching the surface."

Key Takeaways:

That’s a wrap.

As always, the journey doesn't end here!

Please share and let us know what you liked or want changing! 🚀👋

Alastair 🍽️.