"SaaS is Dead"
The prerequisites for this had been accumulating for several years:
- In February 2024, Jensen Huang, CEO of Nvidia, said the industry should ensure no one has to program: now you can speak to a computer in plain human language.
- In December 2024, Satya Nadella, CEO of Microsoft, suggested that business applications in the era of AI agents would "collapse": essentially, they will just become databases, and agents will take over the business logic.
- In February 2025, Andrej Karpathy coined the term "vibecoding." He used it to describe a way of writing programs without looking at the code, trusting the model.
- In March 2025, Dario Amodei, CEO of Anthropic, promised that in six months, AI would write 90% of code.
In February 2026, AI agents that independently work with documents and data were added to this, and the market got seriously scared. Analysts at Forrester called what was happening a "SaaS-pocalypse."
What We Agree With
The barrier to entry has indeed fallen. Development, design, copywriting, and intellectual labor in general have become more accessible. A service that previously took months of work and the budget of an expensive product team can today be put together by one person over a weekend.
It got to the point where Sam Altman, CEO of OpenAI, shared that his acquaintances were placing bets on what year the first one-person billion-dollar company would appear.
This is good news. The market will expand: many new products will appear, although most of them will be raw.
What We Disagree With
There will be more products, but that won't increase the number of high-quality ones. A raw service is very different from a product that sustains a business, and the boundary between them is clearly visible.
As long as a service remains an MVP (the first version to test an idea), one person can carry it. As soon as the service starts generating revenue, new tasks emerge:
- regulator requirements — for banks and large businesses, this is a separate, massive layer of work;
- integrations and data — a product does not live on its own, but among other company systems;
- security — according to Veracode, in 45% of test tasks, code written by AI contained a vulnerability;
- support — users write in, things break, and they need to be fixed today, not when there is time;
- development — a plan, priorities, and a product process are needed.
Building a service is now cheap, but owning it is still expensive. It needs to be integrated into the company's perimeter, and clients and users won't just appear on their own: marketing and sales are indispensable. At this stage, one person can no longer cope, and a team is required.
Who Watches the Watchmen
AI writes code, designs interfaces, and drafts texts. But someone has to verify that all of this is correct, secure, and actually needed by people.
Developer surveys show this is not a theoretical problem:
- Stack Overflow, 2025: 84% of developers already use AI, but 46% do not trust the accuracy of its answers, while only 33% do. The main complaint (66%): solutions that are "almost right, but not quite."
- METR, 2025: experienced developers in a controlled experiment took 19% longer to solve tasks with AI than without it, even though they themselves were confident they were working faster.
It is most dangerous when a product works "almost right." Only someone who understands how it should be can spot the error. Therefore, expertise will not disappear; it will become the watchman.
The product team of the near future will be composed of experts who deeply know their respective fields and intersect at the right points:
- development — architecture and the right principles, without which the product will fall apart in its second year of life;
- user — how people work with the product, what they feel, and where they stumble;
- business — what the client pays for and how the product makes money.
Roles will change: less routine, more verification and decision-making. But AI will not replace depth of understanding; it amplifies it.
"Product teams are moving to a new level of abstraction. By creating digital products, they get the opportunity to think more about architecture, business, value, solution quality, and marketing, rather than about individual variables, frameworks, lines of code, pixels in mockups, and backlog tasks. But when it comes to truly serious issues, who will you trust with them? Will you hand over product security to a vibecoder or an experienced security engineer? And the accessibility interface to them as well, instead of an experienced UX designer?"
Oleg KachalinCo-founder of Rocket Tech
Where the Value of the Product Lies Now
Previously, development was expensive, complex, and consumed most of the resources. This in itself was a barrier: whoever had the money and people for development had the product.
Now this barrier is lower, and what AI cannot do for people comes to the forefront:
- finding an economically viable idea that clients are willing to pay for;
- forming a differentiator: how the product differs from competitors;
- properly communicating the product to the market;
- turning the product into a business.
AI here is a tool that cannot replace people. Jensen Huang thinks the same. In February 2026, at the height of the software stock sell-off, he called it "the most illogical thing in the world": AI agents will use existing software, not replace it. If you have a screwdriver, why invent a new one?
Forrester came to the same conclusion from a different angle: the core of enterprise systems isn't going anywhere. Simple "single-function" services that are easy to replace are under threat. Products with unique data and processes deeply embedded in the company's operations will survive.
Why We Look at Chains, Not Functions
Here we see the main opportunity. The 2025 MIT NANDA report showed that 95% of companies do not get a measurable return from generative AI pilots. The reason is not the models, but the implementation: AI is poorly integrated into real processes.
We observe a similar problem with clients: almost everyone is currently in the experimentation phase. A chatbot here, a text generator there. Automating a single function changes almost nothing: the bottleneck simply moves to the next step.
That is why Rocket Tech focuses on the AI automation of entire process chains in the corporate sector: taking blocks of work from start to finish, rather than individual functions. We described how this looks for us internally in the case study "How to Automate 80% of Routine Work".
"Expertise isn't going anywhere; its role is simply changing. First, you need to deeply understand the client's business and figure out what is actually worth automating. For this, we built our own product, framework, and methodology: a small team of experts, together with AI agents, conducts development at all stages. But people are still responsible for the final product."
Roman NadeinCo-founder and CEO of Rocket Tech
Conclusion
SaaS is not dead; expensive development is dead. Soon anyone will be able to launch their own service, and this is good for the market. But turning a service into a product that sustains a business can still only be done by people who understand what they are building and why. Expertise is not disappearing; its focus is simply shifting from "writing code" to "deciding what code is needed and verifying that it works." And this is only about creation. Selling the product and growing it into a business without an experienced team will also be impossible.
