Anyone can build your product now over a weekend. So where is the moat?

What Happened

You spend 6 to 12 months building a bare bones product. You take that demo out and try to raise $5M. You hire more devs and a product person. Then you raise again, this time to hire the people who will actually sell it, deliver it, and answer the phone at 2 AM when it falls over. 4 rounds, 3 years, maybe an exit.

That script is dead.

Once Anthropic shipped Claude Cowork and Claude Code, investors sat down with a calculator, and roughly $1 trillion of enterprise software value evaporated - about $300 billion of it in a single session. People started calling it the SaaSpocalypse (Forbes). The question the market was asking was pretty simple: if an IT team can vibe code its own internal tool over a weekend, what exactly is that per seat subscription buying? Then, GPT 5.5 landed on April 23 as the Codex default. Claude Code moved to Opus 4.8, then Opus 5. Zhipu dropped GLM 5.2 on June 13 - 753B MoE, 1M context, MIT license - at roughly a sixth of GPT 5.5's cost. Grok 4.5 arrived July 8 trained on real Cursor session data at $2/$6 per million tokens, against Opus 4.8 at $5/$25. Kimi K3 showed up as the strongest open weight coder out there.

4 frontier class coding agents in six months. I can now stand up a real, usable product - auth, database, API, a UI a paying customer will log into - in a matter of days for about $200 in tokens and hosting.

Product is easy now, Service is not

Sequoia's thesis: the next $1T company sells work, not software. Sell a copilot and you compete with every model release. Sell the outcome - books closed, contracts reviewed, claims handled - and every AI improvement widens your margin instead of threatening your product. The insight most people miss: for every $1 spent on software, roughly $6 goes to services. SaaS chased the software dollar. AI chases the services dollar at software margins. Not AI for accountants. The AI accounting firm. Which dollar are you chasing? - Ruben @rdominguezibar

He is summarizing Julien Bek's March 2026 Sequoia essay, "Services: The New Software." Bek's example is the one that got me: a company spends $10,000 a year on QuickBooks and another $120,000 on an accountant to close the books.

We built an entire industry fighting over the $10,000. The ratio is not made up. Global IT services spending runs about $1.73 trillion against roughly $1.43 trillion for software (Forbes).

A health insurer that fired Salesforce

Fred Turner, founder of Curative, went on 20VC and said the quiet part very loudly. His company killed a Salesforce contract after vibe coding an internal CRM in two months.

No one was using Salesforce anymore. We just recently canceled our Salesforce contract because we have an internal CRM that was built, was vibe coded, that is working better, is more integrated into what we're doing. $600,000 a year. Gone to zero. - Fred Turner, Curative (BigGo)

Business Insider confirmed the cancellation notice on July 21. Curative is cutting about 80% of its SaaS spend this year. Google Looker got replaced by a Snowflake native layer built by one or two people. The core claims platform - which they migrated to only a few years ago - is going in house.

And this is a HEALTH INSURER. Regulated, HIPAA, audit trails, credentialing. The exact vertical where complexity was supposed to be the vendor's permanent moat.

But the CRM is not even the interesting part. Curative built an agent called Gwen that does provider network contracting end to end. She takes a lead, Googles the practice, scrapes price transparency files to see what competing payers are paying, finds the contact on ZoomInfo, sends personalized emails, negotiates the rate, redlines the contract by writing Python that edits the Word doc, opens DocuSign and signs it.

The 5 things that flipped

1. Product became the easy part. Anyone can clone your feature set over a weekend now with $200 and a coding agent. Your moat was never the code, it turns out - it was the fact that code used to be expensive. What nobody can clone in a weekend is the migration, the domain knowledge, the trust, and knowing which of the 40 things a client asked for will actually sink the project. If your pitch is "we already built it," YOU DO NOT HAVE A MOAT ANYMORE.

2. We can stop selling everyone the same menu. For 20 years we all designed the same thing - multi tenant platform, finite feature set, sell identical features to every prospect, and anything off the roadmap got a polite "we'll take that to product" and a backlog slot that never came. Customization was expensive. It isn't now. We went from McDonald's, where the menu is fixed and you cannot get the sauce on the side, to having a personal chef - and the chef costs roughly what the menu used to.

3. Square peg, square hole. In our own shop, every client now gets a dedicated private server with a custom software suite built exactly for their operation. Not a tenant on our platform. Their server, their data, their code. Round peg, round hole. Hexagonal peg, hexagonal hole. Weirdly this turned out to be LESS complexity, not more - I am no longer writing feature flags and conditional logic to make one codebase impersonate five products, then explaining to client C why a bug in client A's edge case broke their Tuesday. Client C's code does client C's job and nothing else is in there. The security conversation also got very short: nobody else's data touches this box, because there is no cross tenant.

4. Build the CRM. We built our own instead of renting a finite feature set. The bit I keep showing people - you upload the note from your first meeting with a prospect and it creates the account and the opportunity. No form, no 14 required fields, no "which picklist value is this." The artifact you already produced becomes the record. That is trivial to build now and nearly impossible to get from a platform serving 150,000 customers off one roadmap. And it is not just me and Curative - Greenleaf Management saved about $100K a year (they were paying Salesforce PLUS a full time admin PLUS an outside consultant), Hank AI went from $40K to $500. But this is real at 20-70 employees with a weird workflow. It is not real for a 1,000 person company ten years deep into a Salesforce org.

5. Nobody has time to keep up, and that is the whole business. The average enterprise runs 23 AI tools and only 38% can even inventory them (Larridin). 73% of small businesses say they need help making AI work and nearly half admit they cannot pick the right tool. Four frontier coding models shipped between April and July - a controller at a property management company is not tracking that, and she should not have to. This is why FDEs went from a Palantir curiosity to the hottest job in enterprise software. In May, OpenAI stood up The Deployment Company with over $4 billion and immediately bought Tomoro's 150 engineers. Anthropic, Blackstone, Hellman & Friedman and Goldman announced their own embedded engineering firm aimed at mid sized companies. Accenture launched an FDE practice with Microsoft. Read that again - the model labs, sitting on the highest margin software products ever created, are pouring billions into a SERVICES business. They watched 95% of pilots die between the demo and production and figured out the constraint on their own growth was never intelligence.

The obvious counterarguments

It is early days , and there is a school of thought which believes it is too early to call SaaS dead.

Enterprise customers have all tried to do it themselves, but they realize you can't achieve enterprise-level reliability and security with vibe coding. - Srini Tallapragada, President, Salesforce

If there is a "SaaSpocalypse," it may be eaten by the "SaaS-quatch" because there are a lot of companies using a lot of SaaS because it just got better with agents. - Marc Benioff, February earnings call

ServiceNow points at a 97% renewal rate. Loka CEO Bobby Mukherjee - who has actually gotten clients 40 to 80% software cost reductions doing this - still usually tells them to keep the SaaS, because rip and replace pulls engineering off whatever actually differentiates the business. Turner's Anthropic bill has been sextupling month over month for six months, from tens of thousands to millions a month.

🫤 Dileep's Skeptical Takeaway:

The 2022 playbook was to build the product and then buy your go to market with venture money. In 2026 it runs backwards - the product is nearly free and the go to market, the domain knowledge, the migration, the support and the judgment about what NOT to build is the entire business. "Cheap to build" doesn't mean "Cheap to keep". None of the hard parts have gone away. You still have to sell your product, market it, differentiate it, deploy it, and most importantly - maintain and support it.