Here Be Monsters: What Do You Still Need to Know When Agents Do the Work?

AI coding agents are changing software ownership because they can produce code faster than engineers can absorb it, and may build systems no human has reviewed line by line. A CI failure illustrates the tension: an agent had packed complex components into a single file, triggering a complexity warning. Inspection showed the logic was correct, but the code needed restructuring. The author uses autonomous underwater “divebots” as a metaphor for agents: they can extract value at scale, yet may also get lost, produce misleading results, or create systems that become difficult to maintain. Neither extreme response - reading every generated line nor ignoring implementation entirely - is adequate.

The central argument is that owning a system means having a useful mental map of its abstractions, knowing enough to spot trouble, and being able to investigate unfamiliar parts when needed - not knowing every line or having written the code by hand. Traditional apprenticeship through years of manual coding and debugging may not be economically realistic for new engineers. Instead, the author argues for building ownership through strong fundamentals, broad knowledge, and deeper investigation when a problem demands it. Fundamentals such as algorithms, networking, concurrency, and distributed systems help engineers recognize failure patterns; breadth helps them judge which areas are risky and connect ideas across fields. Together, these skills let engineers direct agents while retaining the ability to diagnose, repair, and explain the systems they produce.