A recent Sequoia Capital essay argues that the next trillion-dollar company will be “a software company masquerading as a services firm.” The core idea is simple: don’t sell the tool, sell the work. Instead of giving customers software to do bookkeeping, close the books for them. Instead of helping lawyers draft contracts, deliver the contract.
There is something important in this thesis, but I think it can go further.
Historically, software succeeded because people preferred the speed, control, and repeatability over services. Software is visible. It is inspectable. It allows people to understand what is happening, intervene when necessary, and adapt the process to their own context. Services, by contrast, often feel like black boxes. You hand work over and hope for the best. Hoping you get the same team or person that was pitched to you rather than it being outsourced to the intern.
The Sequoia argument assumes that AI changes this equation because machines are becoming even more capable of doing the work themselves. That is partly true. But capability alone has never been the only reason people choose software.
The real shift is not from software to services. It is from software that helps people do work, to software that does more of the work on their behalf.
That distinction matters.
For decades, software has been designed around tasks. Accounting software records transactions. CRM systems store customer data. Project management tools track projects. The user still carries the cognitive burden of translating those tasks into outcomes.
AI changes that. For the first time, software can understand the outcome the user is trying to achieve and actively work towards it. The user’s goal is no longer “update a spreadsheet” or “fill in a form.” The goal is “hire a great candidate,” “close the books,” or “renew this insurance policy.” AI allows software to operate closer to those goals.
But that does not mean users suddenly want to surrender control. If anything, the rise of AI makes trust more important than ever.
The reason organisations have historically been reluctant to automate end-to-end decision making is not because rules engines were incapable of handling the common case. It is because the world is messy. Every process contains edge cases, exceptions and contextual factors that nobody anticipated when the system was designed.
Large language models are different because they handle ambiguity remarkably well. They can navigate situations that would have broken traditional automation. They can reason across incomplete information, make sensible assumptions and recover from unexpected situations.
Yet users still know, instinctively, that no system can perfectly understand every edge case. They know there will be mistakes. They know there will be situations where context matters more than pattern matching and systems that can adapt to bring that context along with every decision will win.
Trust therefore becomes the central UX challenge of the AI era.
The best AI products will not be the ones that hide everything behind an “autopilot” button. They will be the ones that allow users to progressively delegate responsibility. They will make recommendations before taking actions. They will explain their reasoning. They will show their work. They will make it easy to inspect, correct and override decisions.
Good AI UX is not about removing humans from the loop. It is about allowing humans to move around the loop.
Sometimes the user wants to perform the task manually. Sometimes they want assistance. Sometimes they want the system to complete the task and simply report back. The winning products will support all three modes seamlessly.
This is where I think the Sequoia thesis is directionally right but incomplete. Customers do want outcomes. They increasingly want software to take responsibility for more of the process. But they do not necessarily want a service. They want a trusted partner that can act autonomously when appropriate and transparently when necessary.
The future is software that understands the user’s goal well enough to behave like a service when asked, while still preserving the visibility, flexibility and control that made software win in the first place.
