I wonder if Sam Altman knew that social media would explode over his offhand remarks about how different generations use AI. What made that statement so compelling was the mysterious notion of using AI like an operating system. Everyone thought: What does that even mean? How can I tap into AI the way these youngsters do?
Personally, I’m not a fan of sweeping generalizations like this. However, I love reflecting on the different ways people use applications, especially when considering how these habits will reshape software development. So here’s what I think it means for AI to be the operating system.
Traditional OS
Before we dive into using AI as an OS, let’s take a step back to take a look at what an architecture at a traditional enterprise looks like.
In the age of the web, the operating system of a modern enterprise is the browser (yes, mobile is a thing, but it mostly built as an extension of browser and not always a big factor for an enterprise). We migrated systems to the cloud, made applications portable and accessible by moving them to the web and broke down capabilities into platform APIs and their backing micro services that use http as the primary interface.
In the age of the web, the operating system of a modern enterprise is the browser
However, even with all of these advancements at the end of the day organizations spend considerable time orchestrating the business. Aside from what a traditional company exposes to their paying customers, there are also hundreds if not thousands of applications that enterprises run. Majority of them require people to manually interact with the application. For example, a sales rep at a large company may need to use a CRM like Salesforce or HubSpot (sometimes multiple instances of both) to pursue an opportunity, which is then manually entered into a different core system for further administration.
The only way business operators can interact with systems in any sophisticated manner is through a user interface. And this is where browser-based applications become a limiting factor. Users can only do what the application allows them to do. A badly designed browser application will easily cripple productivity—we’ve all been there, staring at the screen wondering if it was designed by someone who actively hates us.
The limiting factors of browser based applications force user to be the unwilling orchestrator of the business process.
The other limiting factor is that each application comes with its own interface. This is understandable from a development perspective, but it inevitably leads to business users juggling multiple applications to complete a single end-to-end workflow. Our business user becomes an unwilling orchestrator, following a business process that essentially pieces together different apps to get anything done.
The New OS
I previously wrote about letting AI take ownership of your orchestration layer. While I was initially referring to application code, I want to expand this concept further. Business orchestration—which has traditionally been performed through multiple browser-based applications—can now be handled by AI.
Here’s why this is possible:
- Planning: Given a task and a known set of available tools or other AI agents, an AI assistant can devise a plan for how to complete the task.
- Iteration: It can attempt to perform steps using the provided tools and, based on results, revise the plan and continue iterating until successful.
- Cross-system execution: If multiple systems need to be updated, AI will do so on behalf of the user following established processes.
- Completion notification: Once the task appears complete, it notifies the user who originally made the request.
The business user no longer needs to hop between multiple systems. Their job is to prompt and to validate.
AI becomes the orchestrator, reasoning through business processes and iterating to completion—freeing users from navigating multiple systems to simply expressing intent.
This addresses all the limiting factors we had before:
First, AI applications don’t need to anticipate every possible user need upfront. This makes it much easier to build useful applications that don’t handcuff users with predetermined workflows.
Second, AI applications have a unified interface: natural language. We no longer need to switch between multiple applications. There may still be multiple systems under the hood exposing their operations via AI tools, but from the user’s perspective, it’s all one seamless experience.
Finally, this removes the orchestration burden entirely. AI will orchestrate the use of these tools based on user intent, not user navigation skills.
Traditional UIs become obsolete when AI can generate any view on demand and replace every button click with natural conversation.
Our new operating system also comes with a few additional advantages. Most applications can be reduced to presenting data and offering actions that modify it. Using natural language we can ask AI to search for any record without having to click on filters and perform look ups. The retrieved data can rendered in any way that you want – list, table, file, graphics, audio and if you are really longing for retro in this new world then you can even ask it to be rendered in html. This is as good as it gets for personalization and accessibility.
What’s in the future?
To power this future, we must maintain a strong foundation in APIs. APIs will continue to offer the most efficient way for AI to interact with core systems. Admittedly, even this area is likely ripe for optimization, but I think that comes later in the transformation cycle.
My advice to all teams remains the same: maintain a strong focus on APIs. However, instead of building traditional UIs, think about exposing your APIs through AI tooling. This represents the true democratization of APIs that enterprises have been seeking for years.
When AI becomes the interface, every employee becomes a power user. No more training sessions on yet another internal tool. No more fighting with clunky enterprise software that feels like it was designed in 2003 (and probably was). Just natural language requests that get things done.
This is what using AI as an operating system really means—and it’s going to change everything.