How AI Will Destroy the SaaS Market (And What Comes Next)

By Mizanur Rahman Momin — Founder & CEO, Convo Digital Ltd

Scroll through X or LinkedIn for five minutes and you’ll see the same headline, repeated in a hundred fonts and a thousand threads:

AI is going to destroy SaaS.

As someone who markets SaaS products, sells lifetime deals, builds software with AI coding agents, and runs a community full of SaaS founders and early adopters, I don’t think that headline is wrong. I think it’s lazy. It’s the correct conclusion wrapped around a shallow argument.

Here is the more precise version, and the one I actually believe:

AI is not about to delete the SaaS industry. It’s about to destroy the business model that almost every SaaS company was built on — and the companies that don’t move fast will be destroyed along with it.

That distinction matters enormously. Because if you get it wrong, you either panic and do nothing, or you dismiss the whole thing as hype and keep defending a model that is quietly rotting underneath you.

This post is long on purpose. I want to lay out the full picture: where the SaaS model came from, the exact assumptions AI is breaking, which categories of software are most exposed, who survives, what the replacement business model looks like, how pricing changes, what the counterarguments are, and what I’d actually do if I were running a SaaS company today.

Let’s start from the beginning.

Part 1 — The SaaS era: how we got here

To understand what AI destroys, you have to understand why SaaS existed in the first place.

Before SaaS, software was a product. You bought a license, installed it on your own servers, paid for maintenance, and upgraded every few years when the vendor released a major version. It was heavy. It was expensive. It was slow. But it worked, and it created gigantic companies.

Then, starting around the mid-2000s and accelerating through the 2010s, a new model took over. Instead of selling software, you’d rent it. Instead of installing it, you’d log into it through a browser. Instead of paying once, you’d pay every month, per user, forever.

We called it Software as a Service.

And it was a genuinely better model — for a while. Why?

Because it aligned everyone’s incentives just enough. The customer avoided huge upfront costs and got continuous updates. The vendor got recurring revenue and a reason to keep improving the product. The cloud made it technically possible to serve millions of users from a shared infrastructure, and multi-tenancy made it economically efficient to do so.

By the 2010s, the entire industry had weaponized this into a playbook. It went something like this:

Find a repetitive task that businesses or individuals hate doing.
Build thin software that makes that task faster, cleaner, or more organized.
Charge per seat, per month, priced against the salary of the person doing the task.
Land and expand — get one team in, then spread across the company.
Stack features every quarter so the product feels like it’s always getting better and churn stays low.
Grow through content, SEO, and paid acquisition, then defend the lead with a brand.
Raise capital, grow at all costs, optimize for the metric the market rewards.

This playbook minted some of the most valuable companies on earth. Project management, CRMs, HR software, accounting, customer support, analytics, marketing automation, design tools, developer tools — an entire civilization of “software that helps you do your job.”

And for a long time, the moats were real:

Build cost was brutal. Writing software required expensive, specialized people and years of iteration. That kept competition scarce.
Switching costs were high. Migration, retraining, and integration pain made customers sticky even when they were unhappy.
Distribution was hard. Getting attention and signups required money, brand, or a decade of SEO.
Data was trapped in schemas. Your structured database was a genuine competitive advantage because structuring data was hard.

Every one of those advantages is now eroding, and they’re eroding at the same time. That’s what makes this moment different from every previous “X will kill SaaS” panic.

Part 2 — The five assumptions AI breaks

SaaS rests on five quiet assumptions. Most founders never wrote them down, but every decision in the industry is built on them. AI breaks all five.

Assumption 1: Building software is expensive

The whole pricing structure of SaaS — monthly subscriptions, per-seat fees, annual contracts — exists because building and maintaining software is expensive and risky. You’re recouping a large capital investment over time.

AI breaks this. A capable general model plus a coding agent can now produce, in hours, what used to take a team weeks or months. I’m a non-technical founder and I ship products with AI coding agents, Next.js, and Supabase. That sentence would have been a joke five years ago. Now it’s my Tuesday.

When building is cheap, the thing you built stops being the valuable part. The value moves to whatever surrounds it: distribution, data, trust, and workflow ownership.

Assumption 2: Software features are the product

SaaS sold discrete capabilities: summarize this, categorize that, generate this report, tag these records, draft this email, extract this data. Each capability was a feature, and features justified the price.

AI breaks this. Almost all of those capabilities are now a prompt. Summarization, classification, extraction, drafting, tagging, translation, reporting — these aren’t products anymore. They’re default behavior. The “AI-powered” tier sellers were charging for six months ago is now a sentence inside a chat window that costs fractions of a cent to run.

When features become free, feature-based pricing has nothing left to stand on.

Assumption 3: Humans do the work

The seat-based model is the cleverest part of SaaS. You price per user, which means your revenue scales with headcount, which means you capture a slice of the labor payroll of the very function you’re software-izing.

AI breaks this. Agents don’t need seats. When one AI agent does the work of five users, your revenue mechanically drops even as customer value increases. You end up charging for people who never log in. That’s not a pricing bug — it’s a business model collision.

Assumption 4: Switching costs protect the incumbent

Retention in SaaS has always been at least partly artificial. Migration is painful. Data is trapped. Teams are trained on your UI. It’s easier to keep paying than to leave.

AI breaks this. When rebuilding the alternative takes a weekend, migration pain shrinks. Data export is table stakes. And increasingly, the “alternative” is just a prompt that speaks your customer’s own data. Switching costs collapse precisely when build costs collapse.

Assumption 5: Structured data is a moat

The real product behind many SaaS tools was a well-organized database with a nice workflow on top. Structuring messy reality into clean rows and columns was genuinely hard, and that difficulty was protection.

AI breaks this. Models can now read chaotic inputs and produce structure on demand. Your schema isn’t a moat anymore — it’s an input. The question becomes: do you own data that can’t simply be handed to a model and recreated elsewhere?

Part 3 — Erosion of every moat, in detail

Let’s go one level deeper. Because “AI destroys SaaS” isn’t a single event. It’s five or six erosion processes running simultaneously, and each one is boring on its own but devastating in combination.

The build-cost collapse

The most important change is that the cost of a credible software product has fallen, and keeps falling. This has three consequences:

Competition explodes. If anyone can build your tool, many will. The niche you won for yourself gets crowded in months.
Feature parity becomes trivial. Your differentiation evaporates as competitors clone your best features instantly.
The provider becomes a competitor. The model providers and AI platforms increasingly ship the exact features thin SaaS apps charge for. You’re building on land someone else owns and can fence off.

This is the classic “commoditize your complement” dynamic. When AI commoditizes the building of software, the software itself becomes a complement — and complements get squeezed.

The prompt-ification of features

Take any legacy SaaS category and ask: what is this tool fundamentally doing?

A transcription tool? Speech-to-text plus formatting — now a model call.
A journaling or note app? Summarization and organization — now a model call.
A resume or document editor? Generation and rewriting — now a model call.
A basic analytics dashboard? Aggregation plus natural-language explanation — now a model call.
A “research” tool? Search plus synthesis — now a model call.

When the core competency is a model call, you’re not selling software. You’re reselling intelligence with a markup, and there are a hundred sellers of the same intelligence.

The seat-pricing collapse

Seat pricing has a hidden assumption: work scales with people. AI inverts it. Work scales with compute, and compute is getting cheaper.

Think about what happens to a customer support platform priced per agent, when AI resolves most tickets. Either:

The customer shrinks their support team → your seat revenue collapses, or
The customer keeps the team but stops seeing why they’re paying per seat → churn.

Either way, the pricing model fights the value you deliver. The companies that survive this will be the ones who re-price before their customers force them to.

The wrapper reckoning

The clearest casualties are wrappers: thin apps whose only real asset is a prompt, a UI, and a payment link. These boomed because they were fast to build and benefited from novelty and SEO. But a wrapper has no floor. There’s no data advantage, no workflow depth, no integration gravity. It’s a distribution race, and distribution is where wrappers lose to platforms.

The knowledge commoditization

Many SaaS tools were really lookup and formatting services: “we know where the information is, and we present it nicely.” That’s exactly what models do natively. When knowledge access is free, the tools that monetized access to it must find something else to sell — or die.

The distribution inversion

Here’s a subtle one. In the SaaS era, you built a product and then acquired distribution. In the AI era, distribution is often the only moat left, because the product is easy to copy. That flips the whole strategy. It means brand, community, embeddability, and being the default inside a workflow matter more than feature checklists.

Part 4 — A risk map: which SaaS categories are most exposed

Not all SaaS is equally threatened. Here’s how I think about the risk tiers.

Risk tier	Category profile	Examples (generic)	Why
Extreme	Single-function, prompt-replaceable, priced per seat	Summarizers, classifiers, simple generators, “AI writing” tools	Core function is now a model call; zero moat
High	Thin workflow wrappers, information middlemen	Basic research tools, template generators, light CRMs	Easy to rebuild; value is formatting/lookup
Moderate	Multi-feature tools with some workflow depth	Project management, basic analytics, HR tools	Retention partly artificial; automation removes seats
Low	Systems of record, deep integrations, network effects	ERPs, payment rails, core infra, collaboration hubs	Switching costs and data gravity are real
Inverted	Tools that become more valuable with AI	Data platforms, orchestration, security, evals	AI increases their usage rather than replacing them

The mistake most founders make is assuming they’re in the “low” tier because their customers are loyal. Often, that loyalty is switching cost wearing a costume — and switching cost is exactly what’s collapsing.

Part 5 — Who dies first

Let’s name the profiles. If you recognize your company below, you have a strategy problem, not a marketing problem.

1. The thin feature wrapper

One capability, sold monthly, easily replicated. It survived on novelty, SEO, and being first. AI made “first” meaningless and “replicable” fatal. Diagnosis: no floor.

2. The per-seat repetitive-work tool

Any tool whose value is “a human sits here and does repetitive work.” The moment that work automates, the seats disappear. Diagnosis: your pricing is your enemy.

3. The information middleman

Tools whose main job is to find, organize, or format information that a model can find, organize, or format for free. Diagnosis: your product is a lookup.

4. The artificial-retention incumbent

Products that stay because leaving is painful, not because staying is great. When migration becomes trivial, these bleed. Diagnosis: your moat was friction, and friction is gone.

5. The “we added AI” incumbent

Companies that bolted a chatbot onto a legacy product and called it a transformation. This is the most dangerous position of all, because it feels like progress while keeping the underlying model intact. Diagnosis: cosmetic change, structural rot.

6. The bloated and slow

Large teams, heavy process, slow shipping. In a world where a solo founder with agents ships weekly, being slow is a death sentence, regardless of how good your architecture is. Diagnosis: speed is the only edge and you don’t have it.

Notice the pattern: the companies that die are the ones protecting a business model instead of solving a problem.

Part 6 — Who survives, and actually wins

Now the useful half. The survivors are not the biggest SaaS companies. They’re the ones whose fundamentals AI strengthens rather than erodes.

1. Owners of proprietary, compounding data

Not data you can scrape from the web. Data your customers generate by using you — outcomes, corrections, preferences, labeled results — and data that improves your product over time. This is the single most durable moat left. Scratch data is everywhere; compounding data is rare.

2. Owners of workflow and trust

AI makes answers cheap. It does not make decisions cheap. Somewhere in every serious workflow there’s a human who has to approve, sign, ship, or be accountable. If you own that moment — the approval, the audit trail, the compliance layer, the system of record — you own the value chain. Trust does not commoditize the way prompts do.

3. Systems of record and integration gravity

When AI can build the tool, the winner is whoever holds the data, the integrations, and the switching cost. ERPs, payment rails, and collaboration hubs survive because they sit at the center of everything and their value increases as clever tools plug into them.

4. Companies that price on outcomes

If you charge per lead, per dollar saved, or per ticket resolved, then automating human work doesn’t hurt you — it helps you. Your revenue decouples from headcount and reattaches to value. This is the single most important strategic move a SaaS company can make right now.

5. AI-native architecture

Not “AI features.” AI-native — where the model is part of the runtime, not a bolted-on tier. These companies can iterate faster, price differently, and absorb model improvements automatically instead of rewriting their roadmap every quarter.

6. Distribution and community owners

If the product is easy to copy, the audience isn’t. Founders with a real audience, a real community, and real trust can launch faster and defend better than any feature set. (This is why I spend so much time on community and personal brand rather than another landing page.)

The through-line: AI destroys tools and rewards outcomes.

Part 7 — The new model: from software-as-a-service to service-as-software

Here’s the shift that actually matters, the one the headline is gesturing at.

Old model: you sell a tool. Your customer hires (or already has) people to use it. Your price maps to seats, and your value is convenience.

New model: you sell the finished work. Software does what people used to do. Your price maps to outcomes, and your value is results.

I call this service-as-software — the collapse of the line between the tool and the labor the tool was built to augment.

What changes when you adopt it:

Pricing moves from seats to usage or outcomes. You stop charging per person and start charging per result.
Onboarding stops being a 40-step tour and becomes “connect your data, then let it run.”
Retention stops depending on the friction of leaving and starts depending on the results you deliver.
Product stops being a dashboard you stare at and becomes an agent that acts, with you supervising.
Marketing stops selling features and starts selling outcomes: not “AI-powered analytics” but “we found you $X in wasted spend.”

This is uncomfortable for existing SaaS companies because it means voluntarily cannibalizing seat revenue. But the alternative is having a competitor do it to you. The companies that survive are the ones willing to eat their own revenue before someone else does.

Part 8 — Pricing is the whole game

If I could give a SaaS founder one piece of advice for the next three years, it would be this: your pricing model is more exposed to AI than your product is.

Let’s walk through the options.

Seat-based pricing (per user / per month)

The problem: revenue scales with headcount. AI removes headcount. You get punished for delivering value.

Where it still works: collaboration tools where value genuinely scales with participants (the network is the product), and systems of record where each seat represents real, non-automatable authority.

Usage-based pricing (per API call, per task, per credit)

The upside: revenue follows workload, so automation volume can grow revenue.

The risk: if models get cheaper, your per-unit price collapses too, and you may not capture the value you created. Usage pricing without a value anchor can race to zero.

Outcome-based pricing (per result, per dollar saved, per lead)

The upside: you decouple from headcount and attach to irrefutable value. This is where the smart money is going.

The challenge: attribution and measurement. You need to prove you caused the outcome. That requires instrumentation most SaaS companies don’t have yet — and building it is a real strategic investment.

Hybrid pricing (platform fee + usage + outcome bonus)

The realistic answer: most survivors will run hybrids. A base platform fee for access and trust, usage pricing for volume, and outcome bonuses at the top. It hedges model-price collapse and rewards results.

The strategic point stands: if you can’t explain how your revenue grows when your customer automates more, you have a pricing model that AI will eventually bankrupt.

Part 9 — The data moat, properly understood

Everyone says “data is the moat.” Almost everyone is wrong about what that means.

There are three kinds of data, and only one of them is defensible.

1. Public and scrapeable data

If your dataset can be recreated by crawling the web or licensing a public source, it is not a moat. It’s a starting point. Models already have it, and your competitors can get it too.

2. Private but static data

Data you have that others don’t — but that never changes and never improves. This is a temporary advantage. It buys you time, not a moat. Someone will eventually assemble a comparable set, or a model will generalize past it.

3. Proprietary, compounding data

This is the real moat. Data your product generates through use, that gets better over time, that reflects real outcomes and real corrections, and that feeds back into making the product smarter. Every user makes it more valuable. Every month makes it harder to copy.

The strategic question is therefore not “do we have data?” but: “what does our product learn, from our customers, that no competitor can learn without them?”

If you can’t answer that, you don’t have a data moat. You have a database.

Part 10 — Distribution becomes the moat

When the product is easy to build, the product stops being the advantage. So where does advantage go? To distribution.

In the AI era, distribution takes forms most SaaS companies underinvest in:

Community. A real audience you own, that trusts you, that shows up. This is nearly impossible to clone and nearly impossible to buy.
Embeddability. Being inside someone else’s workflow, so you’re a default rather than a decision.
Integrations and ecosystem. The more surfaces you connect to, the higher the cost of leaving.
Brand. In a sea of identical capabilities, brand is how a buyer chooses when they can’t evaluate the difference.
Speed. The ability to ship weekly means you can ride trends and absorb model improvements faster than incumbents.

This is why I spend so much of my time building a community and a personal brand instead of endlessly polishing product pages. In a market where everyone can build the same thing, the person with the audience wins.

Part 11 — What happens to the SaaS market itself

Zoom out from individual companies to the industry. Several things happen at once.

Consolidation and acqui-hiring

As thin products lose their floor, many will be acquired for their users or their team rather than their tech. We’ll see more acqui-hires and more “we’re joining the platform” shutdown notices. This is already visible; it will accelerate.

Commodity deflation

Categories that were once priced in dollars per month will be repriced toward the cost of compute plus a thin margin. Writing assistants, summarizers, and light generators are already drifting toward free. This deflation is deflationary for revenue even where usage grows.

Valuation repricing

Markets eventually reprice businesses whose moats turned out to be switching cost and whose pricing turned out to depend on headcount. The multiple compression in parts of SaaS is not a sentiment problem; it’s a fundamentals problem finally being recognized.

New category creation

Everything that dies frees demand for what replaces it. We’re seeing new categories form around orchestration, evals, agent security, data ownership, and outcome measurement — the picks-and-shovels of the new model. Some of today’s most important companies will come from these, not from the old SaaS playbook.

The name survives, the substance changes

The category will keep calling itself SaaS while the actual model becomes something else — usage-based, outcome-based, agentic, data-anchored. Names are sticky. Business models are not.

Part 12 — The counterarguments (where “AI destroys SaaS” is wrong)

Intellectual honesty demands I argue the other side. The thesis is directionally right, but it overshoots in specific places. Here’s where.

“AI will destroy all SaaS” — false

Systems of record, infrastructure, security, and network-effect products are not destroyed. AI increases their importance. Saying “AI destroys SaaS” as a blanket claim ignores that a large chunk of SaaS is the plumbing AI runs on.

“AI makes software free” — false

AI makes some software cheap and some features free. Durable software — reliability, security, compliance, integration, support, and accountability — stays expensive. Nobody wants a free tool that loses their data or gets their company fined.

“Features don’t matter” — oversimplified

Features are no longer the moat, but they still matter for winning the initial trial. The point isn’t that features are worthless; it’s that they’re no longer defensible. Defensibility moved elsewhere.

“Incumbents are doomed” — mostly false for the strong ones

A well-run incumbent with data, distribution, and trust can adopt the new model and win. The threat is not incumbency; it’s complacency. Incumbents have what startups lack: customers, data, and trust. If they move, they can be extremely dangerous.

“The market shrinks” — likely false

The SaaS market may shrink in per-unit price while exploding in total usage. Software will do far more work, for far more of the economy, at much lower unit cost. Revenue moves from seats to outcomes, and the total pie can grow even as old line items disappear.

So the correct thesis isn’t “AI destroys SaaS.” It’s: AI destroys the SaaS business model, spares the durable parts of the industry, and rewards whoever adopts the new model fastest.

Part 13 — A rough timeline (my best guess)

Predictions age badly, but they force clarity. Here’s how I think this plays out.

Now → 1 year:

The wrapper reckoning accelerates. Many thin AI tools shut down or get acquired.
Seat pricing starts breaking visibly in customer-facing categories.
“AI features” stop being a differentiator and become table stakes.
A wave of new AI-native products launches from tiny teams.

1 → 3 years:

Outcome and usage pricing become normal, not exotic.
Data moats separate winners from losers decisively.
Consolidation intensifies; acqui-hires become routine.
Systems of record get more valuable, not less.
The first generation of purpose-built “service-as-software” companies reaches meaningful scale.

3 → 5 years:

The default way businesses buy “software” is by buying results, with agents doing the work.
The old per-seat model looks as dated to that generation as desktop licenses look to us.
A small number of data-anchored platforms dominate; the long tail of thin tools is gone.
The label “SaaS” persists, but the substance is unrecognizable versus 2015.

I’m not certain of the timing. I’m fairly confident of the direction.

Part 14 — What I’d do if I were running a SaaS company today

Enough theory. Here’s the playbook I’d run.

Audit every feature against a general model. Which parts of your product does a model now do for free? Be brutal. If the answer is “most of it,” your problem is strategy, not pricing.
Name your proprietary data. What do you learn from customers that nobody else can learn? Build the roadmap around making that data richer and more central.
Redesign your pricing for an automated world. Move toward usage or outcomes. Ask the hard question: when my customer automates their work with my product, does my revenue go up or down? If it goes down, fix it before a competitor does.
Own a workflow end-to-end, not a step. Features are copyable; complete loops from trigger to result are much harder to displace.
Become AI-native by architecture. Put the model in the runtime. Build so that better models automatically make you better, not so that every model upgrade forces a rewrite.
Invest in distribution like it’s your only moat — because it nearly is. Community, audience, embeds, integrations, brand. These survive copying.
Ship like a small team, even if you’re not one. Speed is the only edge that compounds in a commoditized market.
Build the instrumentation for outcome pricing. If you want to charge for results, you need to prove you caused them. Start measuring now.
Cannibalize your own revenue on purpose. Better you than the competitor who will do it anyway.
Fall in love with the customer’s outcome, not your software. This one is a mindset, but it drives all the others.

Part 15 — What this means for everyone else

This isn’t only a founder issue.

If you’re a buyer: the next three years are your best negotiating window in decades. Massive capability at collapsing prices. But watch vendor lock-in and data portability — with thin vendors disappearing, your data’s fate matters more than ever.

If you’re an investor: the question is no longer “is this SaaS growing?” It’s “when this customer automates, does revenue rise or fall, and what non-copyable asset does this company own?”

If you’re a marketer (my world): you will sell outcomes, not features. “AI-powered” is already meaningless. What sells is provable results, real data, and trust. The skills that matter — positioning, proof, community, and relentless shipping — only become more valuable.

If you’re an employee: the seat-based world rewarded headcount. The new one rewards judgment, accountability, and the ability to direct agents. The people who thrive will be the ones who own outcomes and wield AI as leverage — which is exactly why I build with it instead of fighting it.

Part 16 — The honest take

Let me land this where I started.

AI won’t “destroy SaaS” in one dramatic collapse. It will do something quieter and more brutal: it will starve the old model of oxygen while a new one grows in the exact same market. Seat-based, feature-based, wrapper software slowly stops making sense. The category keeps its name while the substance underneath it changes completely.

The companies that die are the ones that protect a business model instead of solving a problem. The companies that win are the ones that accept the obvious truth this whole moment keeps proving:

People never wanted your software. They wanted the result. AI just made the middleman optional.

That’s not the end of SaaS. It’s the end of selling software for its own sake — and the beginning of selling outcomes at scale. The founders who understand that early won’t just survive the destruction. They’ll be the ones holding the market when everyone else has finished grieving their old pricing model.

Adapt, or get automated out of existence. There is no third option.

Key takeaways (TL;DR)

AI doesn’t delete the SaaS industry — it destroys the business model most SaaS companies are built on.
The moats that protected SaaS (build cost, switching cost, features, structured data, hard distribution) are eroding simultaneously.
Thin wrappers, per-seat repetitive-work tools, information middlemen, artificial-retention incumbents, and “we added AI” bolters are most exposed.
Survivors own proprietary compounding data, workflow and trust, systems of record, distribution, or outcomes.
The new model is service-as-software: selling finished results, priced on outcomes, not seats.
Pricing is more exposed than product. If your revenue falls when your customer automates, you have a time bomb.
The counterargument is real: durable software, systems of record, and network-effect products survive and thrive.
People never wanted your software. They wanted the result.