Silicon Valley programmers shift from coding to managing AI agents

In short: Many programmers are doing less hands-on coding and more managing of AI agents that can write, test, and ship software.

What's going on

Software teams in Silicon Valley are increasingly using “AI agents,” which are AI tools that can carry out tasks in steps, like a digital junior employee. Instead of writing every line of code themselves, engineers give the agent a goal, review what it produces, and decide what to do next.

These agents can handle routine work such as writing code, running tests (automated checks to see if the code works), and deploying code (putting the software update live for users). Tools from companies like OpenAI, Anthropic, and Google, plus newer startups, are being used to take on whole work tickets with minimal human input.

Because of that, the job is shifting toward clearer instructions and stronger oversight. Engineers are spending more time breaking business problems into step-by-step requests the AI can follow, checking quality without digging into every implementation detail, and coordinating several agents across different departments. Industry data in the report says 90% of organizations use AI agents for coding, 86% use them on production code, and 42% let agents lead development with a human reviewing the work.

What to watch

Hiring is starting to favor people who can delegate well, think strategically, and communicate clearly with AI systems, not just people who type code quickly. Leaders quoted in the report say humans are still needed for judgment, creativity, and final decisions, especially when the AI makes mistakes. A key question is how widely companies will trust these agents as they move from “helpful assistants” to “digital employees” working across more of the business.

Source: NYTimes