Summary

Maor Shlomo sold one-person, bootstrapped Base44 to Wix for $80M and doesn’t regret it even with Harry citing revenue now past $100M: the deal pays him on revenue milestones, and joining Wix — same audience, “the best marketing team in the market,” lean product team kept “startup-y” — “basically triples my chances at actually building something special” versus bootstrapping against competitors raising huge rounds.
His core call: vibe coding becomes the largest software category, with CRM, support, and project management folding into it — “there would be a time, might be in a few months, might be in a few years, where it will be easier to build your own Salesforce type of CRM than to actually buy a license.” Harry’s pushback was blunt: “With the greatest of respects, that’s wrong… And then do maintenance and updates. Are you serious?”
On the no-moat critique (Benioff, Cannon-Brookes, Matt Prince, Dylan Field and Carl Pei have all spun up Base44 equivalents): “It’s relatively easy to create a vibe coding tool. It’s very very very hard to create a platform” — the moat is a vertically integrated mini-cloud (built-in database, auth, integrations, scheduled tasks, not Supabase like the rest of the industry) serving apps that run to millions of lines of code.
He isn’t worried about Lovable, Replit or Bolt — “Google. By far.” If Gemini decisively wins the model race, Google has the compute, the full stack including Google Cloud, the data and the suite “to build an incredible empire,” and conquering vibe coding is the next logical move for any runaway model winner.
The LLM market is nothing like cloud: “you change one string in the code and you move millions in spend to a different provider” — Base44 shifted complex-application and large-context workloads to GPT-5 thinking/pro, and would shift overnight to Gemini 3 if it’s better. His warning for Anthropic: a rival beats Sonnet 4.5 for six months and revenue sees “such a fast decline, but also they had such a fast ascend.”
On the margin critique: “from all things in the industry, the margins are the least thing that I’m worried about” — margins are decision-dependent, a prompt router to cheap/open-source models can improve them drastically, and strategy already assumes “the prices of model will go down towards zero.” The Composer playbook — grow fast at bad margins, then swap in your own trained model — is one Base44 is considering, potentially fine-tuning open source on its “billions, almost trillions lines of code.”
Positioning calls: Anthropic ($360B) over OpenAI ($500B) “every day” — faster-moving, coding-focused, B2B-first; Cognition ($12B) over Cursor ($29.3B) because this is not a winner-take-all market; and against Lovable’s reported $6B, “Wix is very much undervalued today.” No AI speed bump: “even with existing models we’re only scratching the surface of economic value.”

Deep dive

1. The $80M sale: a bootstrapper’s three-way junction

Base44 started “just for fun” — Shlomo wanted to get back to coding after a capital-heavy first company — then traction went vertical: 10,000 users, 100,000 users, and the business became very profitable. That left three paths: stay bootstrapped and alone and risk “getting eaten by companies that are raising a ton of money,” raise aggressively, or find a parent company.
He resisted selling at first, telling Wix CEO Avishai: “If [the company name is unclear in the captions] was sold to Microsoft, probably there wouldn’t be any of it today” (the raw name is unclear; likely Canva is the referent). What flipped him: same target audience, “the best marketing team in the market… aggressive, creative, absolutely awesome,” and a structure where the product team stays lean and startup-y while Wix absorbs support, legal, and operations.
No regret, for two reasons: the deal compensates him on revenue targets and milestones (he can’t discuss the $100M+ revenue figure Harry cites — public company), and the whole point was a shot at the biggest category: “Without it, my chances would be really really small.”

2. “Easier to build your own Salesforce than buy a license”

The thesis: existing categories — CRM, support systems, task and project management — “are going to fold into the vibe coding category,” and a time is coming, “might be in a few months, might be in a few years,” when building your own Salesforce-type CRM beats buying a license. Harry’s pushback, worth keeping in full: “With the greatest of respects, that’s wrong. You’re not going to have people build their own Salesforce CRMs… And then do maintenance and updates. Are you serious?”
Shlomo’s refinement: not one-prompt CRMs but “software becomes more liquid” — templates and open-source starting points that users then vibe-code into their own shape (“I want this version in Arabic from right to left”). The buyer logic: the code is yours, the data is yours, no vendor lock-in, no feature bloat.
The origin story carries the argument: his wife, a tattoo artist, needed a CRM; configuring drag-and-drop vendors was “hell on earth. And I’m like a software person, right? I should know how to do that.” Base44 exists because an LLM could build something leaner and exactly fitted — no enterprise features, no team-management settings she’d never use.
Who dies: “every software that’s basically front end on top of a database with some team features and some integrations” and no moat, plus — a decade out — every software company that needs implementers. Salesforce itself might survive: its sheer size is a moat, “Benioff is a smart person,” but the category’s companies get smaller.

3. The moat rebuttal: tools are easy, platforms are hard

Harry lists the founders who’ve built Base44 equivalents in three months — Benioff, Mike Cannon-Brookes, Matt Prince, Dylan Field, even Carl Pei. Shlomo’s answer: “It’s relatively easy to create a vibe coding tool. It’s very very very hard to create a platform that could help people build products that are functional, complex enough for real world use cases” — some Base44 applications run to millions of lines of code.
His dismissal of demo-ware is the episode’s sharpest line: “‘Hey look, I’ve built this monday.com clone with a single prompt.’ No, you didn’t. It’s going to take you hundreds or thousands of prompts to get to something you’re going to use day to day.”
The moat is infrastructural: “You’re almost building like a small cloud” — databases, user management, built-in integrations, scheduled tasks, an ecosystem the LLM can reach into. Websites and landing pages will commoditize; complex functional apps won’t — and they’re the profitable segment, because website builders stop prompting once “okay, this looks pretty,” starving a credits-based business model.

4. Liquid software: incumbents embed, pro tools stay up-market

Incumbents won’t just be eaten — they’ll build vibe coding into their platforms: Monday Vibe on monday.com’s boards and permissions, Microsoft on Office, and Shlomo imagines Salesforce as a system of record where “you can create the UI that you want.”
The pro tools (Cursor, Cognition) own the developer who wants code, terminals, and enormous enterprise projects — his example of what Base44 can’t do: a firewall, native, performant, shipping logs to a centralized server. Base44’s internal north star is “what’s the percentage of software that we can build today” — pushed forward by adding scheduled tasks and built-in AI agents integrated with WhatsApp and Slack, “so now you can actually build bots, not only applications.”
Both layers will drift toward each other — pro tools adding visual builders, app builders extending downward — but the users are fundamentally different: “either you serve the developer… or you serve the non-technical user, which is completely different.” For Base44’s lawyer-building-a-law-firm-system user, code “is not what they want to do.”

5. Fear Google, not Lovable — and the switching dynamic that breaks the cloud analogy

Asked to rank Lovable, Replit and Bolt, he demurs (Replit is “more technical,” hardcore devs go to Cursor) but drops the real answer: “I’m not worrying about those three.” The risk is a model provider winning “by a wide margin — then the next logical thing is to conquer the vibe coding market.” Who? “Google. By far. They have the compute. They’re moving very fast” — plus Google Cloud, the data, and the suite “to build an incredible empire.” He’s “hoping for a tough race, and it looks like that right now.”
The switching mechanics, as told: Base44 split workloads between Sonnet and Gemini, moved to Claude 4/Sonnet 4 (“the best designer,” best at writing a project from scratch), then found GPT-5 thinking and pro “really good at solving hard bugs” and shifted complex, large-context workloads to OpenAI — “you’re literally moving hundreds of thousands of spend with the simplest change.” If Gemini 3 is better, “which is what people are nowadays saying,” spend moves overnight.
Harry’s synthesis — worth keeping: everyone analogizes LLMs to cloud, but “you do not switch your cloud provider with a line of code in seconds multiple times a month.” Shlomo agrees from scar tissue: his previous company’s cloud migration was so painful “if I’d known how tough it was going to be, I probably wouldn’t do it.”
The Anthropic stress test he poses: someone ships a better, stronger, cheaper model than Sonnet 4.5 and holds the lead six months — “what happens to Anthropic’s revenue? … This is such a fast decline, but also they had such a fast ascend.”

6. Margins “suck today” — and why that’s his smallest worry

Harry: “Today they suck.” Shlomo: margins are decision-dependent — “you can change it from a low margin to a very high margin with just one decision: how great of a service you want to give your users.” The fix is a router: “change the color of the button from blue to red” doesn’t need “the heavy Sonnet maximum thinking” — forward it to a smaller, cheaper, faster model and margins “improve drastically.” One caveat aimed at Replit-style 20–30-minute autonomous runs: “you could burn a lot by not understanding what the user actually wants.”
The strategic bet he and Avishai settled on: “we’re taking into consideration that the prices of model will go down towards zero” — so Base44 deliberately doesn’t staff cost optimization, even where optimizations are visible today.
The Composer play is his template for the category: grow “as fast as you can” on expensive frontier models, win distribution (“every developer probably installs it”), then migrate users to your own efficient trained model — “overnight, your margins get tremendously better.” Base44 is thinking the same: potentially fine-tuning open source on the “billions, almost trillions lines of code” inside the platform.

7. The weeks-to-copy game and the backend bet

Echoing Fiverr founder Micha’s line to Harry that time-to-copy collapsing from years to weeks is the biggest change in software, Shlomo confirms: every feature gets copied in weeks or months — their “app suggestions” feature was cloned in days. “It’s fine. That’s the game that we’re playing now” — velocity, delight, UX decisions, taste.
The exception is the founding bet: a fully built-in backend — database, user management, auth, integrations, analytics — built in-house rather than on Supabase or Neon “which is what the rest of the industry is using.” Replicating it, and migrating millions of users off Supabase, “is going to be very hard” — a bet they’re “deepening further and further.”
His success metric is unusual: minute-to-minute sentiment analysis of tens of thousands of chat prompts — users talk to the agent like a human (“Fantastic job, this is exactly what I wanted” versus “This is not what I wanted”). The jump from Sonnet 4 to Sonnet 4.5 showed up as a visible improvement. The metric itself has shifted: bugs are now rare (“the LLMs are really smart, the infrastructure is battle tested”) — the question is whether the agent does what was asked without breaking other features.

8. The angel lens: vertical integration or get eaten

On Benchmark GP Av Randhawa’s growth-over-margin framing, he mostly agrees: “the game for our category is growth… there’s almost no person that shouldn’t be a user.” But leanness compounds — the lesson from his first company, which was very capital-heavy—$130M and 100+ people: “when you’re lean and you’re a sane business, that does really good to all other aspects” — especially when model costs are “90%, 99% of the spend” and repricing every few months.
On today’s headline ARR numbers: the bar has moved — his first company’s $2M ARR in eighteen months earned cheers; now showing the hockey stick to $10–20M changes everything, though when he calls 0-to-10-in-a-year “insane,” Harry pushes back: “there’s quite a lot that do 0 to 10.” His two real tests: will it be eaten by model providers, and is it vertically integrated?
The eaten test: if the value is mostly prompting plus UX — his example, an AI photo editor — “I don’t think that’s going to be a healthy business for the next decade,” with Jasper AI the cautionary tale. He passed on Manus-type agent startups despite aggressive growth curves: “I’m not sure there is a lot behind it rather than prompting and giving it tools that are for sure going to be part of ChatGPT or Gemini.”
Where nobody’s investing but should: unsexy industries, end to end. Rather than just build a Harvey competitor — “build a law firm… buy a hospital, optimize everything on every layer, make it AI-native, and maybe in a few years be the first to introduce robots as surgeons.” Five years ago that was “the most bad idea you can think of”; the platform shift makes it the interesting one.

9. The valuation card: Cognition over Cursor, Wix over Lovable, Anthropic over OpenAI

Cursor at $29.3B or Cognition at $12B? “Probably Cognition” — betting on Cursor is betting winner-takes-all, and “our notion from the previous age of venture capital where winner takes it all is wrong… everybody will take a niche.” VC “is not necessarily a winner takes it all but a huge returns type of market” — which is why he entered against viral Bolt and Lovable in the first place.
Forced by Harry to choose — Lovable at a reported $6B and Replit at three-to-four (“I’m sure they’ll raise at six soon cuz they’re kind of memetic”) versus Wix — “my very non-objective opinion is that Wix is very much undervalued today”: there’s a moat in being big and in a marketing organization with experience. He won’t call the startups overvalued without seeing their internals — he tracks their revenue and published metrics, but not their internals — and Wix taught him “focus on your business, focus on a great product,” versus his first company where “40% of my mind was occupied with what competitors are doing.”
OpenAI at $500B or Anthropic at $360B? “Anthropic every day. Moving faster, they cracked something in coding, very focused on delivering value to B2B customers. They don’t have the first-mover advantage, but they have better models and they’re better positioned for growth.”

10. No speed bump, delusional labs, and the 95% future

On the bubble: the question is whether AI is a platform play — model improvements accruing to the whole ecosystem — or a knife-fight where model companies’ margins erode. “Me personally, I don’t believe we’ll hit a speed bump… even with existing models we’re only scratching the surface of economic value.” His proof-by-existence: “if a one-person team can get sold for $80 million… you’ll be able to do a lot more stuff with less people,” and that value trickles down to cheaper, better products.
What frontier labs are delusional about: “the moat that they have. They don’t have what they think they have” — one better model or lower price craters revenue, and “the Chinese players are now changing the entire market dynamics.” The mitigation he credits: building up the stack, like Claude Code — even if Gemini 3 beats Sonnet, “there’s going to be a huge number of people staying with Claude Code because they’re used to it.”
Quick-fire: Benioff and Tenev’s 50% AI-written code goes to “95%. Over 100% for sure — you’ll be prompting agents” in two years; Base44 is already ~90% because he structured the repo for LLMs to navigate. His advice to non-technical builders: build for your own problems (“the biggest hack”), and embrace disposability — “build a first version that you know you’re going to throw out… you’re not working with a human,” so revert freely and iterate.
And on geography: not being in Silicon Valley doesn’t hurt — the talent war there is brutal, Lovable is thriving from Europe, and “Wix is a monster giving us their backs… that’s my Silicon Valley.”

The 2045 Superintelligence Timeline: Epoch AI’s Data-Driven Forecast Coca-Cola: The Complete History & Strategy (Audio)