Better agent memory often starts with a smaller task

A lot of agent systems lean on memory to patch over unclear workflow boundaries. This piece argues that smaller, better-defined tasks usually improve reliability, cost, and debuggability more than adding another memory layer.