So precise a measurement of the year, of course, would have been irrelevant to an early farmer, even if the astronomers could have measured it. A shift of five or ten days won't really matter too much, since annual variations in the weather will be greater than that. Once the calendar is out by a month or more, however, problems will arise if the farmer depends upon the calendar. It doesn't necessarily follow, however, that an agricultural society needs to keep its calendar tied, even loosely, to the seasons. Even if we assume that farmers really need someone to tell them when spring is coming, an alternate procedure would be for a specialist, for example a calendar priest, to announce when the proper planting time fell in the calendar.
If you are a farmer, an arbitrary period has its problems. Of particular interest is the best time to plant crops—an activity tied to the seasons, which are in turn linked to the earth's revolution around the sun. If all you have is an arbitrary cycle, it's much more difficult to tell just when you should plant. Some agricultural societies, therefore, have tried to link their calendar to the length of the time it takes the earth to rotate around the sun. Of course most agricultural societies have not known that the earth orbits around the sun. From the perspective of an earth-bound observer, this motion (along with the tilt of the earth's axis relative to its orbit around the sun) translates to the changes in position relative to the horizon where the sun rises (or sets) each day. Over the course of a year, this position will change, day by day. If you look East at sunrise, the point at which the sun appears to rise each day will move over a year between a maximum north and south position (the solstices). The midpoint of this motion, through which it passes twice each year, is the equinox. If you live within the tropics, at some point in the year the sun will pass directly overhead (i.e., at noon a straight stick will cast no shadow). The length of one complete cycle is known as the mean tropical year. In 1900 CE this period was 365.24219878 days, and the length of the tropical year has been shown to decrease by 0.0053 seconds per year (a value which becomes important when we start talking about 1000-year intervals).