“Just Make It SaaS” May Be the Worst Startup Advice of 2026
Pure software is no longer automatically a premium business. AI is eating the old SaaS multiple, and founders need to build where software alone is not enough.
There is a particular sentence that should make every founder slightly nervous:
“We have a successful business, but investors want us to make it SaaS.”
I heard a version of that today from a startup. And I understand why investors say it.
For fifteen years, SaaS was the golden animal of venture capital. Predictable revenue. High margins. Clean dashboards. Expansion revenue. Annual contracts. A pricing page with three tiers and a button called “Talk to Sales,” which usually means “we are about to charge you based on how expensive your shoes look.”
It was beautiful. It was also very financeable.
So when investors see a messy, working business, their instinct is often to push it toward SaaS.
Package it. Standardize it. Put it in the cloud. Charge monthly. Add onboarding emails. Create a dashboard. Call it a platform. Everyone claps.
The problem is that this advice may now be dangerously outdated. Not always. There will still be great SaaS companies.
But “make it SaaS” is no longer automatically strategic. In many cases, it may be the fastest way to move from a defensible business into a software category that AI is about to commoditize with the cheerful violence of a wood chipper.
The SaaS Multiple Is in Trouble
My fear is not that SaaS companies will disappear. They will not.
Software is still useful. Businesses still need tools. People still need systems, interfaces, compliance layers, workflows, data, permissions, and someone to blame when Salesforce becomes a theological event.
But pure software is losing some of its magic.
The old SaaS premium came from scarcity. Building good software was hard. Hiring engineers was expensive. Distribution was difficult. Integration took time. Workflows were sticky.
Once a company adopted a SaaS product, replacing it was painful enough to require a committee, three consultants, and a heroic lack of imagination.
AI changes that. It makes software easier to build. It makes internal tools cheaper. It makes workflow automation more accessible. It lets small teams produce what used to require product, design, engineering, documentation, and customer success all marching in a neat little parade.
This does not kill SaaS. But it does compress the valuation story.
Some companies that used to dream of 20x revenue may start being valued like normal businesses. Three times EBITDA. Possibly with a grim little spreadsheet attached. And when that happens, investors will not be calm and philosophical.
They will behave like someone discovered a rat in the fondue.
The Problem With Pure Software
Pure software used to be enough. Now pure software increasingly needs a moat.
Not a pretend moat. Not “we have AI in the roadmap.” Not “our UX is delightful.” Not “we are vertical-specific,” which often means “we changed the nouns in the onboarding flow.”
A real moat. Data rights. Distribution. Regulation. Hardware. Physical operations. Network effects. Trust. Workflow ownership. Deep integration. Domain expertise.
Or something that AI cannot simply regenerate, imitate, or wrap inside another interface.
If your product is just software sitting between a user and a task, you should ask one unpleasant question:
What happens when the user’s agent can do this directly?
Because that is where we are going.
A lot of SaaS was built around humans clicking through processes. But agents do not want dashboards. Agents want APIs, permissions, goals, constraints, and execution paths.
If your product is designed only for human operators, and the next buyer is a human’s agent, your interface may suddenly look like a Victorian kitchen in a robotics factory.
Charming. Not ideal.
Where I Would Look Instead
If I were building or backing startups now, I would be more interested in areas where software is not the whole product.
First: strong offline and hardware businesses.
Hardware is no longer the graveyard it was treated as for years. Robotics, sensors, edge devices, industrial tools, smart infrastructure, logistics, agriculture, healthcare devices, energy systems, and physical automation all feel more alive now.
Why? Because AI needs bodies.
It needs sensors. It needs actuators. It needs real-world interfaces. It needs things that move, measure, inspect, build, harvest, clean, deliver, and repair.
The physical world did not become less important because software got smarter. It became the next frontier.
Second: agent-native commerce and services.
Most online commerce and service design assumes a human buyer.
A person searches. A person compares. A person reads reviews. A person clicks checkout. A person books a call. A person forgets the password and ruins everyone’s day.
But what happens when the buyer is an agent?
Your personal AI finds the supplier, compares terms, negotiates, checks compatibility, books the service, monitors delivery, and complains with perfect grammar.
That changes everything: Product discovery changes. Pricing changes. Reputation changes. APIs matter more. Machine-readable trust matters more. Human persuasion matters less at the first layer.
The companies that redesign commerce, procurement, booking, support, logistics, and services for agentic customers may be far more interesting than another dashboard for humans to pretend they enjoy using.
Third: token optimization for the enterprise.
This one is not glamorous, which usually means it is important.
Agentic systems are about to make token usage explode. Not by 20%. Not by “some efficiency concerns.” By orders of magnitude.
When companies move from chatbots to autonomous agents, they stop paying only for answers.
They start paying for loops. Planning loops. Tool calls. Memory. Retrieval. Evaluation. Retries. Parallel agents. Verification. Monitoring. Long-context workflows.
The CFO will eventually look at the invoice and make a noise normally heard only from injured wildlife.
Enterprises will need token auditors. Model routers. Workflow optimizers. Loop controllers. Cost observability. Agent budget enforcement. Caching layers. Eval-aware routing.
Systems that decide when to use a frontier model, when to use a smaller model, when to stop, when to compress context, when to split tasks, and when the agent is simply burning money while appearing productive.
That is not a feature. That is an enterprise survival layer.
The Better Question
So when someone says “make it SaaS,” I think the better question is:
Why?
What is the actual defensibility?
What does AI make cheaper?
What does AI make more expensive?
What part of the value must remain physical, trusted, regulated, embodied, proprietary, or deeply integrated?
Who is the future user: a human, a team, a company, or an agent acting on their behalf?
And what happens to your business when software itself becomes easier to generate than investor updates?
SaaS is not dead. But the lazy SaaS premium is.
The next great companies may still contain software, of course. Almost everything does. But they may not look like classic SaaS.
They may look like physical systems with AI inside: Agent-native marketplaces. Infrastructure for autonomous workflows. Cost-control layers for model usage.
Hybrid businesses where software is the nervous system, not the product brochure.
The old venture advice was: turn it into software.
The new advice may be: build where software alone is not enough.