Those who can attribute themselves to the oldfag days can recall that a cameraman on GTV could make or break the experience for thousand viewers watching the show. The human touch behind the camera was not an ability, but a necessity before the automation helped with followkiller and followpowerup commands. The human would concentrate on the proper point of view and the human would select which battles or areas would be more spectacular for the viewer. Naturally, the human element came with errors that sometimes were simply debated by viewers, sometimes irritating, and at times even irreversible costly: missing a huge quad run or a flag cap, or leaving the camera staring at a ceiling for 2 hours at WCG.

The aforementioned automation made life easier, and at times safe, but what it also did – it took the soul element out of camera work. Today, the camera always follows the killer, always follows quad and flag carriers. Inherently, those are correct choices for focusing the POV, but it comes with a cost at another end – overcorrection. Overcorrection destroys the spectator experience by making games from anything that is soulless, to completely unwatchable.

This spectrum is very easy to dissect:

1. Loss of singularity.
Handing the POV to next player breaks the story. One player game, fueled by his decisions and performance, is one full storyline. Switching the POVs at times becomes something relative to cable surfing, where you are too bored with one good movie to watch, you want to substitute the boring gaps with TruTV top 10 dumbest robbers.

2. Loss of challenge.
Switching POV to a winner seems like a proper thing to do, but why? Let’s say cypher got killed while his opponent is fully stacked. Following cypher will tell the story how to perform a champion-grade rebound and how to overcome the uncertainty and the challenge of the incoming pressure. Viewers get to experience the imminent danger glooming around because killing cypher is an emotional event after all and his way back to control is no less interesting than the killer’s pursuit.
What’s the last time you saw the quad player walking in the room with all that danger, the unchallenging power, dismay, and ultimate destruction? Or maybe, the ultimate success of killing a quad? You haven’t. You get to follow the quad now and get always the one-sided story.

3. Loss of continuity.
Imagine CTF championship game of 8 players, 4 aim beasts on each team, racking up on average 400 kills in 20 minutes. Add to that 3 power ups, and 2 flags. You get to follow it all. This becomes completely unwatchable experience as the camera switches faster than it does the worst episode on MTV2 reality show.

4. Loss of essence.
All of the above converge to the main drawback of soulless automation – the camera does not capture what the game is largely about. One player thought process and strategy. Whether it is duel mind games, TDM team flow, or CTF positioning, auto-follow destroys all of that. Br1ck and whaz might have had the best defense and offense in CTF of all time, you will never know it. You will get to watch their game in 5 seconds intervals mixed with meaningless cess pool in mid.
The issue is even bigger here – failure to capture the inexperienced crowd that needs essence to understand what they are watching. I believe this greatly contributed to popularity of TDM and CTF – people from beginner modes do not get to feel what the game is about when they consider it on stream. Majority of common men fell in love with the game not by playing it, but by spectating <insert idol here>. New players barely get that experience anymore.

As long as the reflection rant lasted, the conclusion is brief:

Everything in moderation:
While the automation is helpful, don’t rely on it fully. Take control of your presentation.

Game modes:
Today, duel manual camera work is doable, in TDM and lolCA it is desirable, and in CTF it is absolutely essential. I do not see much up for discussion here.

Most common source of the game relay today is a single caster. One caster can man the camera and talk about what he sees all at the same time. Have a bind that toggles the follow* variables and execute them only when you feel is appropriate, when you feel that maybe the game is getting boring from current POV or meaningless.