What does it mean that agent evolution versus agent replacement?

Sometimes you should improve an existing agent; sometimes you should replace it entirely.

Example: Your morning planning agent has grown bloated — what started as a 10-minute prioritization routine now takes 45 minutes and includes journaling, email triage, calendar review, and three different inboxes. You could trim it back to its core function (evolution), or you could retire it entirely and build a new agent from scratch that handles planning in a fundamentally different way — say, a single question ('What is the one thing that makes everything else easier?') instead of a multi-step process. The right answer depends on whether the agent's core architecture is still sound or whether it has drifted beyond repair.

Try this: Pick one agent you currently run — a habit, routine, decision framework, or mental model that feels sluggish or unreliable. Write two columns: 'Evolve' and 'Replace.' Under Evolve, list specific modifications you would make to restore or improve it. Under Replace, describe what a fresh agent designed for the same purpose would look like if you started today with no history. Compare the two columns. Which path produces a better agent with less total effort? Notice whether your gut reaction and your rational analysis point in the same direction — if they diverge, the sunk cost fallacy may be at work.