Software engineers' new job isn't writing code — it's designing the boundaries AI agents can't break

The role of software engineers is evolving, moving beyond merely writing code. Modern data platforms have shifted, with tools like Cursor and Claude Code enabling agents to generate distributed streaming pipelines and complex API integrations. These agents can navigate repositories, write test coverage, inspect stack traces, and propose refactors, often preceding engineers in open-source file exploration.

This development raises questions about the future of software engineering. Are we moving towards an industry of rubber-stamping pull requests, or has the work shifted towards more abstract, high-level tasks? To understand this, we can draw inspiration from thermodynamics, which provides a framework for directed work, feedback, loss, and system coherence.

When an agent receives direction in the form of a prompt or business requirement, it transforms that direction into code, tool calls, queries, tests, and system changes. However, every agent loop has losses - it can develop stale assumptions, treat old migrations as current behavior, or accumulate its own history. Human intervention helps by introducing new information, such as failing tests, precise data contracts, or deterministic tools.

Without such signals, agents may generate motion but fail to convert it into useful work. The infinite monkey theorem illustrates this concept, suggesting that repeated attempts, finite constraints, and feedback can lead to coherent outcomes. Yet, enterprise systems with mutable operational state, third-party APIs, and evolving schemas present a more complex challenge.

The 'three-body problem' of enterprise logic aptly captures this complexity. Changes in one system can ripple through others, leading to semantic mismatches and incorrect outcomes. Thus, the software engineer's role has transformed into designing equilibrium - establishing boundaries that make the agent's mistakes visible, specific, and recoverable.