Strategy
SaaS Is Dead. I Buried It in 15 Days. Here's the Proof.
SaaS Is Dead. I Buried It in 15 Days. Here's the Proof.
Last month I looked at the invoice from Intercom and something inside me said "enough."
$132/seat/month. Plus $0.99 for every AI response. We have tens of thousands of students. You do the math.
Pipedrive was no better. Starts as "affordable," climbs to $79/user/month once you add the extras. Combined annual bill for both: somewhere between $60K and $100K.
But the real pain wasn't the money.
The real pain was this: none of these tools knew our students. They didn't know our most common problems, our education consultants' procedures, which student drops off after which lesson. They couldn't measure teacher-student compatibility. They couldn't use the behavioral data flowing from our Insider CDP.
We were paying nearly $100K a year. In return, we got generic tools that didn't know us, built the same way for every company, speaking nobody's language in particular.
It's like living in a rental apartment. Same floor plan, same furniture, same wall color. Want to customize? Upgrade to the Enterprise plan. Hire a consultant. Sign an annual contract.
One night I sat down and thought: "What if I built my own house?"
And I built it. In 15 days.
What Can You Possibly Build in 15 Days?
Now you're probably thinking exactly that. I would have thought the same thing before I started. Honestly, if someone had told me six months ago that we were building our own CRM, I would have smirked and told them to focus on their core business.
But building software in 2026 is nothing like 2020. Andrej Karpathy called it "vibe coding," you might remember. His February 2025 tweet got 4.5 million views. A year later, he updated the term himself: he now calls it "agentic engineering." You give tasks to AI agents, they write the code, you supervise.
That's exactly what I did.
What I built is not a prototype. It's a production system currently serving thousands of students and teachers:
Omnichannel inbox - WhatsApp, Instagram DM, Gmail, webchat, phone... all on a single screen. Our support reps no longer toggle between 5 different tabs.
AI agent system - A 9-step orchestration. It detects customer segments, classifies intent across 17 categories, scores leads using 20+ signals, assesses churn risk across 9 dimensions. Three different Claude models work together: one classifies, one generates content, one makes complex decisions. On top of that, the system takes initiative on its own and uses our MCP tools. It's not just an AI that "responds" - it's an AI that "solves problems."
Autonomous churn prevention - When a student quietly starts drifting away, the system notices. It sends personalized WhatsApp messages. Hour 0, hour 24, hour 48, hour 72. No human intervention.
Self-improving pattern learning - At 2:00 AM every night, a pattern learning process runs. It analyzes the last 7 days of results. It learns what worked. It forgets what didn't.
Over 100 automations. 15 trigger events times 20 action types (and we'll keep expanding).
This is not Intercom. This is not Pipedrive. This is something neither of them could do together. Because this system knows our DNA.
This Is Bigger Than One CTO's Story
So is this just a "look what I built" post?
No. This is part of something much bigger.
In the first week of February 2026, an earthquake hit Wall Street. $285 billion evaporated in 48 hours. Salesforce shares dropped 38% since the start of the year. The media called it the "SaaSpocalypse." In February alone, an estimated $1 trillion in market value was wiped from the software sector.
Where did this panic come from?
From companies starting to realize something. That they can build their own software now.
Klarna eliminated 1,200 SaaS tools. Replaced Salesforce and Workday. Their AI chatbot took over two-thirds of customer service. $40M in annual savings. Revenue per employee jumped from $400K to $700K.
Blinkist replaced $60K/year in SaaS spending with apps built using Lovable and Replit. And the people who built them weren't engineers. Over 20 non-technical employees built their own internal tools.
According to Retool's 2026 report, 35% of builders have already replaced at least one SaaS tool with a custom build. 78% plan to do more this year.
And now Flalingo is part of this story.
This isn't a trend. It's a wave. And the wave hasn't even hit the shore yet.
An Economist Described This 80 Years Ago
Joseph Schumpeter introduced "creative destruction" in 1942. That unstoppable process that continuously transforms the economic structure from within, tearing down the old and creating the new.
SaaS itself was a product of this process. In the 2000s, it destroyed on-premise software. Server rooms shut down, CDs went in the trash, everyone moved to the cloud. SaaS companies were the disruptors of that era.
But disruptors get disrupted too.
Clayton Christensen called it the "Innovator's Dilemma." Large companies become so dependent on their existing customers' demands that they can't see the change coming from below. Salesforce can't abandon per-seat pricing because their entire revenue structure depends on it. Intercom keeps charging $0.99 per AI response because their margins require it.
Both know they need to change. Neither can.
Satya Nadella said it plainly in December 2024. "Business applications are essentially CRUD databases with business logic. In the agent era, this logic will migrate to AI agents."
When Microsoft's CEO says this, it's worth listening. :)
Sam Altman added in July 2025: they've been cutting inference costs by 10x every year for the past five years. He predicted that by the end of 2026, software that used to take a team a year to develop could be produced for just $100-$1,000 in inference costs.
I proved that prediction right this month. In 15 days.
The Jevons Paradox, 1865
William Stanley Jevons discovered that making steam engines more efficient wouldn't reduce coal consumption. The opposite happened: as efficiency increased, the range of applications expanded, and total consumption grew.
The same thing is happening with software. When AI makes software development 10x cheaper, we won't produce the same amount of software. We'll produce 100x more. Software custom-built for every company, every problem, even single-use software.
Kevin Roose of the New York Times called it "software for one." Software written for a single person.
This is SaaS's worst nightmare. Because the competitor is no longer another SaaS company. The competitor is the customer themselves.
The Numbers Don't Lie
Cursor AI became the fastest company in history to go from $1M to $500M in revenue. $1B+ annual revenue. $29.3 billion valuation. OpenAI, Shopify, and Nvidia use it.
Lovable reached $200M annual revenue in 8 months. 8 million users. 100,000 new apps generated every day. $6.6 billion valuation with just 45 employees.
Bolt went from zero to $40M annual revenue in 5 months. The second fastest-growing product in history after ChatGPT.
In Y Combinator's latest batch, a quarter of applications had codebases that were 95% AI-generated.
GitHub Copilot has 20 million users, and nearly half the code its users write is generated by AI.
Jensen Huang said it back in 2023: "Everyone is a programmer now. You just have to say something to the computer."
That day it sounded like an early prophecy. Today it's reality.
The Honest Part
This isn't easy. Not everyone can do it. At least not yet.
A METR study found something interesting: experienced open-source developers were actually 19% slower with AI tools. But they thought they were 20% faster. A serious gap between perception and reality.
According to Veracode, 45% of AI-generated code contains security vulnerabilities. The maintenance burden is real. The "last 20%" problem is real.
Someone on Hacker News rightly wrote: "Once you finish building that SaaS replacement, you're also signing up to operate, maintain, and secure it for as long as you run it."
True. But that objection applies to amateurs.
A CTO building contextual bandit pattern learning and multi-model AI orchestration isn't doing "vibe coding." They're doing what Andrej Karpathy defined in February 2026: "agentic engineering." No vibes. Engineering discipline.
And the real question is this: are the risks of a system I built myself greater? Or is the greater risk paying $100K+ per year for a generic tool that doesn't know its students by name?
I Backed My Thesis With My Wallet
This isn't just an engineering story. I made the same move in my own investment portfolio. I exited SaaS stocks. Moved from the application layer to the infrastructure layer. Companies like DigitalOcean and Cloudflare.
The logic is simple: if every company is going to build its own software, demand for the infrastructure that software runs on will multiply. In the 1849 California Gold Rush, the biggest winners weren't the people digging for gold - they were the ones selling shovels and tents.
Same story, different century.
This is not investment advice. But it is an indicator of how deeply someone believes in their own thesis.
SbY: Software by You
I'm proposing a term: SbY. Software by You.
SaaS sold you a rental apartment. SbY is building your own house. AI is the power tools placed in your hands.
Foundation Capital calls this "Service as Software" and frames it as a $4.6 trillion opportunity. Bessemer says "Vertical AI is the new SaaS." DHH is bringing back the "pay once, own forever" model with his ONCE brand.
I'll put it more simply: software will now be built by the person who knows their business best, for their own needs, with AI's help.
And that person will most likely be you.
Where Are We on the Map?
Economist Carlota Perez has an elegant model. Every technological revolution passes through five stages: irruption, frenzy, turning point, deployment, maturity.
SaaS was the deployment phase of the internet revolution. It had matured. Growth was slowing. Prices were rising.
With AI, we've entered a new frenzy. Cursor is valued at $29 billion, Lovable at $6.6 billion. Speculative capital is flooding in. SaaS stocks are collapsing while AI stocks climb.
This frenzy will calm down someday. And in the deployment phase that follows, what's dismissed today as "vibe coding" will become standard business practice.
Just like the people who said "you're going to put data in the cloud? Are you crazy?" in 2005 now run everything on AWS.
The Wind Doesn't Ask Permission
I built a CRM and customer support platform in 15 days.
But the important thing isn't 15 days.
The important thing is that the paradigm shift that made this possible isn't going back.
Every year inference costs drop by 10x. Every year AI tools become more capable. Every year the "buy or build" equation shifts a little more toward "build."
The average company spends nearly $5,000 per employee per year on SaaS. It wastes $21 million annually on unused licenses. Some of that money can now go toward AI tool subscriptions and a few weeks of development time.
The SaaS era taught us a lot. But like every era, it's ending too.
And Schumpeter's wind of creative destruction doesn't ask permission. You know, that "unstoppable process that continuously transforms the economic structure from within, tearing down the old and creating the new."
Hayreddin Tüzel
CTO & Co-Founder @ Flalingo
Best online English learning platform
Tens of thousands of students | And now a company that builds its own software