The End of SaaS and the Era of Liquid Software

For forty years software was built on scarcity. Code was hard to write, so whoever wrote it could charge rent on it. That era is over.

Every year I write a letter to the Vendasta team about the year ahead. This year I told them 2026 is a year of re-founding. We are rebuilding the ship while it is at sea.

I have acted on six big technology shifts since 1989. Each one followed Amara's Law. People overestimated it in the short run and underestimated it in the long run. This is the seventh. The earlier shifts were about distributing software and getting access to it. This one is about generating it.

The wrapper economy is collapsing

For two decades value sat in the application layer. A company put an interface on a database and charged per user per month. That worked because the interface took a team of engineers, a product manager and months of work.

AI now writes most of that surface logic. A user can describe a feature for a dog-walking business and get the database, the screens and the logic in seconds.

Companies that are only a front end on a database have no moat left. As users find they can build their own version that fits exactly, retention in rigid tools will fall. I expect value to go to two places: the model providers who supply intelligence, and the platforms that supply data, capabilities and security. We are the second kind.

Vibe coding is for everyone

Andrej Karpathy coined "vibe coding" in early 2025 for a developer guiding AI in plain language instead of writing syntax. I think it becomes the main way software gets made, for everyone, developers included.

A developer says: "I use AI to build bigger applications faster."

A florist says: "I built a system that tracks my roses, emails my supplier in Holland when stock is low and texts my customers on their anniversaries."

The florist is the market. Business owners who will never see a line of code and badly want software that works the way they think.

Solid and liquid

Solid software is version 5.2. It has a fixed set of buttons. If you need a new one, you file a request and wait six months for 6.0.

Liquid software keeps flowing. AI rewrites it continuously to fit the user and the moment.

Picture a plumbing business that starts doing HVAC. It does not buy an HVAC add-on. The owner tells their AI employee, "I'm doing HVAC now. I need to track refrigerant and compliance." The fields, the logic and the workflows appear. The owner feels like they gave an instruction to an assistant.

When code is disposable, lock-in by feature set disappears. What stays is context: the data, the capabilities and the history that tell the AI how this business works.

A small cloud for the local economy

If code is free, where is the moat? A pure vibe-coding tool starts empty. No customers, no data, no reputation history, no payment rails, no security. An empty shell.

The moat is almost a small cloud, built for the mechanic, the dentist and the realtor.

1. A system of record that gathers and maintains the business's data.
2. The capabilities to do the work across their whole stack: write content, send campaigns, answer the phone, book appointments, send invoices.
3. Distribution through partners with deep knowledge of local business.
4. Identity, permissions and security for the business's staff.
5. Payments, so the business can bill its own customers.
6. Hosting that is secure by default.

A user can generate code anywhere. They cannot generate their data, its history or the ability to act on it.

The engineering gets harder

Over the last month I went deep into our own stack. One thing became obvious. AI makes creation easier for our users and makes the work harder for our engineers.

When a user asks for a feature, the platform has to provision data, secure the endpoints and route the work to the right model in milliseconds. That is distributed systems engineering of the highest order. Our engineers are moving from placing bricks to designing the physics of the building. They are building the machines that build the machines.

AI does the work

The old paradigm: software helps a human do the work. A spellchecker helps a writer.

The new paradigm: software does the work and the human orchestrates. A person sets the strategy. An AI employee writes the content, posts it and answers the comments.

So most of what we build goes into the doing, and the rest into tools for people to direct it. The interface of 2026 is a place where an owner manages their AI employees.

Every job at Vendasta changes too. Nobody is untouched. Someone great who refuses AI falls behind. Someone good who masters it becomes our next star. Our plan is to upgrade our people so they can put AI to work.

Welcome to 2026

In 2026 we stop selling software. We start delivering outcomes. We stop building tools for people to use and start building the AI that does the work.