Technical Context
I read Sequoia's analysis carefully and it struck me: this isn't just another news release, but a framework that makes it easier to view almost the entire market. Their thesis is simple and very compelling: AI is eating away not just at SaaS, but at the huge services market, which is still dominated by people, processes, Excel, email, and endless manual operations.
To translate this from venture capital speak to engineering terms, we're no longer talking about just another copilot next to an interface. We're talking about a system that doesn't just suggest, but does a whole chunk of work. Not "here's a draft," but "I've closed the books, compiled the documents, sent the status update, and flagged where a human is needed."
I was particularly struck by how Sequoia frames the model shift: they used to sell a tool, now they sell an outcome. Not a CRM for a team of brokers, but a completed insurance policy. Not accounting software, but a closed month. Not a dashboard for lawyers, but a completed case workflow.
And the magic here isn't just in the LLM. I constantly see the same thing in my projects: a model without the right scaffolding quickly descends into chaos. You need routing, validation, memory, access controls, a human-in-the-loop, action audits, a proper AI architecture, and fault tolerance. Otherwise, it's a demo, not a service.
Sequoia also hits the nail on the head regarding sales. SaaS could be pushed from the bottom up, with a low-cost entry and self-service. AI-powered services almost always come from the top down, through trust, a pilot project, a custom rollout, and the promise of a specific result. That's because customers are no longer buying buttons, but delegating a critical process.
What This Changes for Business and Automation
In practice, I'd look at it this way: the winners are those who can package automatable labor into a reliable service. The losers are those who just bolt a chatbot onto their old SaaS and hope it's enough. It won't be.
The most interesting shift is in the product's unit economics. Classic software logic was seat-based: more users, more revenue. An AI service has a different logic: pricing is increasingly tied to the volume of work done, per case, per document, per processed task, and sometimes even to the business outcome.
For architecture, this is a radical change. If you want to build AI automation, it's no longer enough to build an interface and hook up a model via an API. You need to design a system of actions: where an agent can make a decision on its own, where it must ask a human, how it confirms task completion, how steps are logged, how failures are handled, and what to do if the model makes a confident mistake.
I would also highlight the new bar for engineers. Today, the value isn't just in "knowing how to call an LLM." The value lies in turning an unstable, probabilistic thing into a predictable business process. This is where real AI solution development begins, not just a cool weekend prototype.
Sequoia points to good verticals: legal, insurance, managed IT, healthcare admin, wealth management. They all share a common pattern: lots of repetitive intellectual work, expensive human labor, and a weak affinity for existing interfaces. This is fertile ground for AI implementation, if it's done not for hype, but to remove a layer of manual work.
But there's a tough part too. This market isn't about easy wins. You need to earn trust, integrate into live processes, handle compliance, and calculate the cost of errors. That's why I'm less and less a believer in universal "AI employees for everything" and more a believer in narrow systems where you can create an AI agent for a specific function and bring it to an industrial-grade quality.
At Nahornyi AI Lab, this is exactly what we do: we look not at abstract AI, but at the specific labor that can be taken off a team's plate without losing control. Sometimes this involves n8n workflows with agents and checks, other times it's a full-fledged AI integration into a CRM, helpdesk, back office, or internal company services.
This analysis was written by me, Vadim Nahornyi, from Nahornyi AI Lab. I work hands-on with AI automation and AI solution architecture: designing agents, building automations, and testing where AI truly replaces routine tasks versus just creating noise.
If you want to discuss your use case, order custom AI automation, get a bespoke AI agent, or build an n8n automation for your process, get in touch. We'll analyze your project like adults, without the magical thinking.