YurijL

Posted on Feb 5 • Originally published at seeto.ai

Are Quietly Killing Mid-Tier SaaS (and What Survives After 2026)

On paper, that’s a rising tide, but zoom in and you see something different…

Intro — the paradox

If you only look at top-level numbers, SaaS doesn’t look like it’s dying at all.

Global SaaS revenue is growing from about $266B in 2024 to roughly $315B by early 2026, on track to more than triple again by 2032.
At the same time, enterprise software spend overall is heading toward $675B+ by 2024.

But… the huge chunk of that software budget is being redirected toward AI. IDC expects large enterprises (the G2000) to allocate 40% of their core IT spend to AI initiatives by 2025.

The money isn’t disappearing. It’s moving.
And a lot of mid-tier SaaS products are directly in its path.

This is what “LLMs are killing SaaS” really means: not that software dies, but that generic, undifferentiated SaaS gets squeezed from both sides — by AI-native tools on one side and AI-augmented incumbents on the other.

Let’s unpack how that squeeze actually works, with some numbers behind it, and what’s left for founders who still want to build in this environment.

Budgets are being rewired around AI, not around apps

The generative AI software market is forecast to grow from $37.1B in 2024 to about $220B by 2030, a 29% CAGR.
Founders love to say “AI is just another feature,” but CFOs don’t agree — they’re literally creating separate budget lines for it.

By 2025, if 40% of core IT spend in large enterprises is tagged as “AI”, that means many traditional SaaS line items stop being justified as standalone apps and start being questioned:

Why do we pay $X per seat for this, if a copilot can do 80% of the job inside tools we already use?

At the same time, we already have 300+ enterprise tools that have embedded generative AI via APIs or in-product copilots.

Result: the default answer to a new workflow problem is no longer “buy another SaaS” — it’s “can our existing stack + LLM do this well enough?”

If your product is that “another SaaS”, you’re in trouble.

LLMs are commoditizing huge chunks of the SaaS value chain

Every SaaS product, at some level, does a few things:

captures data
applies some logic
presents a decision or output

LLMs are eating the “logic + output” part at an insane speed.

Across B2B software leaders, roughly 42.5% already see Generative AI as “transformative” for development, sales and pricing of software.

What that means in practice:

“Smart” features — summaries, insights, recommendations — no longer feel premium. Users expect them by default.

Interfaces shift from rigid forms to conversational or assistant-driven flows.

The difference between your “analysis” and a prompt pasted into a copilot shrinks to almost nothing.

If your SaaS is basically CRUD + reporting + a couple of fancy charts, LLMs don’t just compete with you — they turn you into a pre-built prompt template.

From 2024 to 2026, the generative AI ecosystem is also exploding in sheer volume. Estimates suggest we may be heading toward tens of thousands of generative AI startups globally, potentially 100,000 if current trends continue. Even if only a small fraction survive, that’s a lot of people trying to compress what you charge $49/month for into a feature, a plugin or a script.

Distribution is shifting from “apps” to “agents inside platforms”

A few more numbers:

Generative-AI tools like ChatGPT hit 100M+ monthly active users in 2023, and the major LLM platforms now process billions of prompts per day.

Where do users start their workflows now?

Increasingly, not in your app:

They start in an LLM chat to brainstorm, draft or analyze.
They use an internal copilot inside Notion, Salesforce, HubSpot, Figma, Linear, etc.
They rely on AI “agents” that talk to multiple systems at once.

From the user’s point of view, your product is just one more API endpoint the agent can hit.

This is brutal for mid-tier SaaS because it collapses your brand layer. If the user never logs into your UI, doesn’t see your onboarding, and doesn’t interact with your pricing page, your negotiating power erodes. You become invisible plumbing.

In that world, whoever owns the agent and the starting point owns the relationship. Everyone else fights for margin in the background.

So is SaaS really “dead”?

No. But a certain type of SaaS is dying:

tools that are thin wrappers on top of public data and generic workflows
tools whose only moat is “we built it first”
tools that can’t convincingly answer: “Why wouldn’t I just ask my copilot to do this?”

Meanwhile, the macro numbers still look good for software overall. SaaS revenue keeps climbing; AI revenue keeps climbing even faster.

What’s being killed is the lazy middle — products that sit between spreadsheets and deep systems of record, but don’t own either side.

How to survive in 2026 if you’re building SaaS

The obvious advice is “add AI”, but that’s not enough. Everyone adds AI. Most users won’t even remember which tool shipped which copilot first.

You need to change where your moat sits.

Become a system of record, not a feature.
If your product holds the canonical version of something important — contracts, pricing decisions, experiment history, market intelligence, risk models — you’re not easy to rip out. LLMs can read and act on that data, but they don’t replace the place where it lives.

Own proprietary data or a proprietary lens.
The gen-AI market will be a $200B+ space by 2030, but most of the raw models will be commoditized.

What doesn’t commoditize as fast is:

long-term customer behavior data
labels, evaluation frameworks, scoring systems
opinionated playbooks for a specific niche

The more your product learns from your users in a way that only makes sense inside your domain, the harder it is to copy.

Design workflows, not just interfaces.
LLMs are great at text and reasoning; they are bad at owning responsibility. A SaaS product that just shows “insights” is easy to ignore. A product that:

captures inputs
routes work
enforces steps
logs decisions

…is a workflow. Agents can help inside it, but they don’t replace it.

Treat LLMs as infrastructure, differentiate above them.
By 2025–2026, using a frontier model API will be as normal as using cloud storage. It’s not your moat. Your moat is:

picking the right model or ensemble
curating prompts, tools and guardrails
deeply integrating into boring enterprise systems no one wants to touch

Most SaaS founders still underestimate how much value there is in plugging AI into ugly internal realities instead of shiny new use-cases.

Be honest about whether you’re a product or a consultancy in disguise.
LLMs amplify both. If your “SaaS” only works with a ton of manual hand-holding, own that and price/position it like a high-touch solution — not a $19/month tool hoping to go viral.

The uncomfortable conclusion

LLMs are not “killing SaaS” in the sense of ending software businesses.

They are killing the illusion that:

you can sit in the middle with a generic product
charge a comfortable subscription
and never worry about being commoditized

Budgets are being rewritten around AI.
Logic and UX are being absorbed by copilots and agents.
Distribution is moving to platforms that start with “Ask me anything…”

What survives on the other side are products that: