The manifestation of imposter syndrome today involves questioning what it means to be a software engineer amidst the noise of agentic engineering. While many focus on the hype of "major model providers," the real power lies in using agent harnesses to explore fulcrum points. In the past year, agent harnesses crossed the "can it be done" rubicon, but capability alone is not the majority of the profession.\n\nHow something goes together is what makes all the difference.

Making software debuggable, maintainable, layered, and composable remains a trick that requires extensive, thoughtful reasoning. Current LLMs do not "reason"; they predict, acting as compressed human knowledge. Even the leading edge of frontier models falls short of true reasoning, as highlighted in research like The Illusion of Thinking.\n\nAs we work with LLMs, we must focus on providing concise data and deterministic validation.

We must also navigate risks like the "lethal trifecta" regarding prompt injection. Ultimately, carefully reviewing, planning, and fixing the seams of software is a critical skill. It is about picking the right abstractions to manage cognitive load and choosing what makes sense for the problem at hand.