The Great SaaS Unbundling Is Over. AI Is Rebundling Everything.

Last week, I watched a founder demo a “game-changing” AI product.

It was slick. Clean UI. Confident pitch. The model responded fast. The demo looked like magic.

And I had one thought I didn’t say out loud:

This isn’t a company. This is a temporary interface to someone else’s intelligence.

That sounds harsh. It’s also the new default.

Because AI has done something startups have never dealt with at this scale:

It made building easier than differentiating.

We’re about to live through a weird era where it’s never been cheaper to start a software company, and never been harder to keep one alive.

The honest shift: the scarcity moved

For the last decade-plus, SaaS lived on a simple advantage: software was expensive to build, so shipping functionality created defensibility. If you were early, you got feature gravity. If you were good, you got seat expansion. If you were great, you got a category.

AI moved the scarcity.

Now code is cheap. Features are cheap. “Smart” is cheap.

What’s expensive is:

• Distribution (who can get attention and trust)

• Workflow ownership (who is embedded in how work actually gets done)

• Context (data, history, integrations, permissions)

• Reliability (repeatable outcomes, not cool outputs)

If you’re building in SaaS right now and you still think you’re competing on features, you are playing last decade’s game.

Why it feels like AI is “killing startups”

People keep asking the same question in a hundred different ways:

“Is AI making startups fail?”

The answer is yes, but not like the movie version. AI isn’t marching into your office and replacing your founding team.

AI is doing something more humiliating.

It’s exposing how thin your value is.

If your product is essentially:

• a prompt

• a UI

• a couple integrations

• an LLM API call

• a workflow that ends right when things get messy

…then you didn’t build a moat. You built a checkout page for a capability that’s getting commoditized in real time.

This is why the “wrapper” debate exists. And why it’s both right and wrong.

The “wrapper” debate is a distraction (but also a warning)

You’ve heard it: “Most AI startups are just ChatGPT wrappers.”

On one hand, that phrase is lazy and usually said by people who haven’t shipped anything in years.

On the other hand, it contains a real warning: if you don’t own something unique, you are renting your business from a platform that has no reason to protect you.

Even Stripe’s leadership has pushed back on the dismissal, pointing to how fast some AI startups have grown and basically saying the “wrapper” insult misses what’s happening. They cited examples like Cursor surpassing $100M in revenue and other tools hitting meaningful ARR in shockingly short windows.

I agree with Stripe on one point: the growth is real.

I disagree on what it means.

Fast growth is not a moat. It’s a starting gun.

We are confusing velocity with durability because velocity is what the last cycle rewarded. This cycle is going to reward the opposite:

Depth. Stickiness. Integration. Trust.

Wrappers can become businesses if they wrap a workflow, not a capability.

If your product owns an end-to-end job, including the annoying parts (approvals, governance, routing, audit, exceptions, edge cases), then you are building something sticky.

If your product exists mostly to show off what the model can do, congratulations. The model will do it better next quarter, and someone else will copy your UI next weekend.

Capital is piling into AI, which makes the middle brutally crowded

The second reason SaaS feels unstable right now is that the entire venture machine tilted hard into AI. Not “AI is hot.” More like “AI is the weather now.”

Crunchbase data shows AI captured close to 50% of global venture funding in 2025, up from 34% in 2024, and that funding grew sharply year over year.

That does two things at once:

1. It creates a flood of startups building similar products at the same time.

2. It raises customer expectations instantly.

The market is getting trained to believe every tool should be intelligent, instant, and cheap. Which is exactly what kills differentiated pricing for shallow products.

And the money is concentrating at the top of the stack. Crunchbase also highlighted that a handful of mega-rounds represented a massive chunk of total funding in 2025.

Translation: it’s a winner-take-most dynamic in infrastructure and foundation models, and a knife fight in applications.

So if you’re an “AI SaaS” founder in the middle, you’re dealing with pressure from both sides:

• Platforms above you bundling capabilities.

• Dozens of lookalike apps beside you competing on price and novelty.

SaaS unit economics just changed, and a lot of people are pretending it didn’t

Classic SaaS had a beautiful lie: once you built the product, adding customers was basically free.

AI breaks that lie.

When intelligence is usage-based, cost scales with usage. Inference shows up in gross margin like a tax that spikes when customers love you.

That’s why “seat-based pricing” is starting to wobble. It’s not dead, but it’s no longer sufficient for many AI-native products because the cost driver is consumption, not headcount.

Here’s the part most founders miss:

If your product feels like an assistant, customers want to use it constantly. If your cost scales with use, you can accidentally build the world’s most beloved margin sink.

This is why you’re going to see more pricing experimentation: hybrid, metered, outcome-based, tiered by power users, and pricing tied to value delivered.

If you don’t match price to cost and value, your company becomes a machine that converts growth into burn.

Enterprise adoption is exploding, but monetization is not automatic

There’s another myth forming: “Enterprises are buying GenAI like crazy, so the market is infinite.”

Enterprises are spending. Yes.

Menlo Ventures reported $37B spent on generative AI in 2025, a 3.2x year-over-year increase.

That’s real demand.

But demand is not the same as vendor success.

Because enterprises are also doing what they always do:

• consolidating

• standardizing

• preferring suites and platforms

• requiring governance

• moving slowly when risk is high

Here’s a clean example: Microsoft 365 Copilot.

Microsoft reported 15 million paid Copilot seats, but that’s only about 3.3% of the estimated 450 million commercial Microsoft 365 users.

Think about what that implies.

If Microsoft, with the most obvious distribution advantage in modern software, is still converting only a sliver of its base into paid Copilot, then monetization is not a given. Value has to be proved. Workflows have to be owned. Change management has to happen. Pricing has to make sense.

So when an early-stage startup tells you “we’ll just land in the enterprise because everyone needs AI,” what they mean is “we have not yet met procurement.”

The new failure pattern: the “feature evaporates” death spiral

Here’s how a lot of AI SaaS companies will die, and it’s going to look confusing from the outside.

1. Launch fast because building is easier now.

2. Get early traction because novelty is a drug.

3. Usage spikes, and so do compute costs.

4. Differentiation shrinks because competitors and platforms catch up.

5. Churn rises because switching costs are near zero.

6. You raise prices to survive.

7. Customers churn faster.

8. Someone bundles your core value into a suite.

9. The company becomes a case study on “timing.”

It won’t be dramatic. It’ll be gradual. It’ll look like “we couldn’t make the economics work.”

Which is technically true.

But it’s usually a strategy problem, not an infrastructure problem.

What wins instead: the five moats that matter now

If you’re building in SaaS right now, you need to be almost paranoid about moats. Not in a pitch-deck way. In a “will this survive an API price change” way.

Here are the moats that actually matter in AI-first SaaS:

1) Distribution

If you don’t have a channel, you are buying attention at auction against 1,000 other “AI” tools with the same tagline.

Distribution is the new feature.

2) Workflow ownership

If you own the messy end-to-end job, you create switching costs that competitors can’t copy in a sprint.

3) Proprietary context

Integrations, permissions, historical data, org knowledge, business rules. This is the stuff that makes outputs reliable and makes your product feel inevitable.

4) Trust and governance

Security, compliance, audit trails, control. This is where cute demos go to die.

5) Outcomes

Stop selling “AI.” Sell the measurable thing that happens because of it: fewer escalations, faster close, higher conversion, lower shrink, better forecasts.

This is why “agents” are the next big battleground. Not because agents are a shiny concept, but because the promise is outcome-driven: software that does work, not software that talks about work.

A contrarian take: “AI-first” is already becoming a red flag

If your whole brand is “AI-first,” you’re late.

In a year, “AI-first” will sound like “internet-first” did in 2005. It’s table stakes. It’s not positioning.

What will matter is:

• Are you category-defining in a vertical?

• Do you own the system of record and the system of action?

• Do you have a data loop that gets better with every customer?

• Can you deliver repeatable outcomes that survive model shifts?

The winners won’t call themselves AI companies. They’ll call themselves the best way to run a business function.

The losers will still be calling themselves “AI-powered” while customers quietly replace them with a feature inside a suite.

How to build like the future is real

If you’re a founder, operator, or investor reading this and wondering what to do with it, here’s the practical filter I’d use right now:

If a platform ships your core feature tomorrow, do you still matter?

If the answer is no, you are not building a company. You are building a demo that will get absorbed.

So build where the platform doesn’t want to go:

• messy vertical workflows

• regulated environments

• systems that require trust

• integrations that take months, not weekends

• domain-specific outcomes that need feedback loops and judgment

And build pricing like an adult. AI costs are real. Your pricing model needs to reflect usage, value, and the cost of intelligence.

The future of startups is not fewer startups. It’s fewer survivors.

AI is going to create more founders than ever because the barrier to starting is collapsing.

It’s also going to create more failure than ever because differentiation is collapsing too.

The startup game is becoming more extreme:

• Launch is easier.

• Copying is easier.

• Bundling is inevitable.

• Attention is scarcer.

• Trust is everything.

If you’re building in SaaS, the mission is simple:

Stop building “AI features.”
Start building workflow gravity.

Because in the next era of software, the product isn’t what the model says.

The product is what the customer can reliably get done.