Old Worlds Shattering
“I’ve never felt this much behind as a programmer,” confessed Andrej Karpathy, Tesla’s former director of AI, in a post on X (formerly Twitter), on Dec. 26, 2025. It sent shockwaves through the tech community. “The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between,” he continued. Karpathy’s words capture a profound moment of transformation—not just for developers, but for the entire digital landscape.
Make no mistake, we are witnessing the emergence of a new technological epoch in which AI agents are fundamentally reshaping how we build, deploy, and interact with software. This isn’t merely an incremental upgrade to our existing tools. It’s a complete reimagining of the programmable layer itself, involving “agents, subagents, their prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations, …” as Karpathy describes. We’re learning to wield what he calls “some powerful alien tool” that comes with no manual, while a “magnitude 9 earthquake is rocking the profession.”
The first realm to feel these quakes? SaaS. Microsoft CEO Satya Nadella, in a remarkably candid podcast with BG2 at the end of 2024, didn’t mince words: “I think, the notion that business applications exist, that’s probably where they’ll all collapse, right in the agent era.” His reasoning cuts to the heart of the matter: Today’s SaaS applications are essentially “CRUD [create, read, update, and delete] databases with a bunch of business logic.” But in an agent-driven world, that business logic migrates to what Nadella calls the “AI tier,” which is a new decisioning layer in which agents orchestrate across multiple databases simultaneously. “These agents are going to be multi-repo CRUD,” he explained. “So they’re not going to discriminate between what the back end is. They’re going to update multiple databases, and all the logic will be in the AI tier. …”
The implications are staggering. SaaS interfaces (such as menus, buttons, and forms arranged for visual consumption) were designed for human interaction. But if AI agents become the primary users of business applications, why maintain elaborate UIs? The agents don’t need to see anything; they need direct API access to accomplish tasks efficiently. The expensive, carefully crafted user experiences that defined the SaaS era become, in Nadella’s vision, largely obsolete.
The New Interface Paradigm
This transformation extends beyond back-end architecture to how we fundamentally interact with technology. The Windows paradigm—with its desktop metaphors, nested menus, and visual hierarchies—is giving way to something more elemental: conversation. The Google-style search bar, once a simple query interface, is evolving into the primary portal for all digital interaction. “Where do you want to start today?” takes on new meaning when that conversational interface can orchestrate complex multistep workflows across dozens of applications. The “I feel lucky” button becomes less joke and more prophecy—a single prompt delegating tasks to a constellation of specialized agents. As technology progresses, we’ll see richer visualizations emerge not as primary interfaces but as control mechanisms— dashboards that help humans oversee and guide agentic systems rather than manually operate software applications.
The Speed of Transformation
The velocity of this change is breathtaking. Google principal engineer Jaana Dogan recently shared a sobering anecdote: Her team spent nearly a year building a distributed agent orchestration system, navigating multiple design paths and internal disagreements. When Dogan described the same problem to Claude Code, the AI generated a remarkably similar system in about an hour.
She noted that the result wasn’t flawless and iterations were necessary, but it resembled Dogan and her team’s earlier work. This isn’t just about coding faster—it’s about democratizing the ability to architect complex systems. Knowledge that once resided exclusively in the minds of senior engineers can now be synthesized and deployed by agents that learn from vast repositories of human experience.