The classic concept of the commander is that it is your avatar on the battlefield. You - the commander - teleport your giant mech suit into enemy territory to build an army from scratch and lead it to victory. This can even be somewhat justified lore-wise by declaring interstellar teleportation to be too expensive to send more than a self-contained one-man army. Commanders from both sides can even enter at the same time if you posit some form of defensive teleportation network that responds to invasions. Just don't ask why they arrive in equal numbers at either end of a symmetric battlefield.
The problem with commanders is that, although they are cool, they go against much of the design of the rest of the game. The dominant, if perhaps not always the majority, view among the developers has been that Zero-K is primarily a game about units. Under this view players are given access to a large toolbox of units and the freedom to mix and match them however they like.
Powerful commanders can limit player freedom by requiring armies to be both built around commanders and built to counter them. But not everyone agreed with this assessment and the mere existence of commanders motivated everyone to try to fix them, so they have had a lot of iteration. So much so that this post will only be able to cover how they work in the early game and early in development.
Our initial commander design came to us from Total Annihilation mostly unchanged: a slow constructor armed with a laser and a gun that dealt infinite damage. It also had some innate income and could cloak, but the main selling point was its highly destructive disintegrator gun, or DGun for short. The DGun let commanders defend themselves against early attacks but could also be used for a wide variety of nonsense. You could build a transport and drop on the enemy to DGun their factory. You could cloak and walk into the enemy army to DGun their striders. You could even DGun enemy commanders, but this would mean death for both of you as these commanders exploded like nukes.
The DGun and nuke explosion made commanders quite volatile. This was counterbalanced by the fact that, since you are the commander, you lose the game if it dies. However, by the time Complete Annihilation got started, a mode called team commends was popular, which meant that a whole team would only lose if they lost all their commanders. This led to fun strategies such as comm bombing, in which you would send your commander on a one-way trip into the enemy front to blow up the whole thing, ideally catching an enemy commander or two in the blast. Comms would walk into the enemy line, often while cloaked, or fly in a transport.
Comm bombs were originally kept in check by a few mechanics that were removed in Complete Annihilation. For example, cloak used to only block vision, not radar, so a slowly approaching radar dot was likely to be a comm bomb. This form of cloak was removed as radar wobble caused it to randomly confer invulnerability depending on whether the dot lined up with the position of the unit. We also removed the special damage type that caused Light Laser Turrets to deal more damage against commanders, the result being that they could fearlessly run into the enemy line. Comm bombing ended up being too swingy and put too much focus on commanders, so we removed it by shrinking commander death explosions. However, this was not even the most controversial change as we also removed the DGun.
The DGun was removed from the basic commander because it had an outsized effect on early army compositions. We were trying the novel idea of giving units roles, of creating things like the Raider/Skirmisher/Riot counter triangle, and the DGun was far too effective at countering riots and assaults. A commander could instantly kill quite an expensive army of short ranged units, to the point that skirmishers dominated the early-mid game just because they could fight from beyond its range. The viability of an army composition depended on how many commanders were left on the other side and on how many commanders you had left to defend it from assault units. The effect of commanders diminished as armies scaled up, but we wanted people to be able to use a larger set of units from the start.
In the post-DGun world commanders are designed to die to about 400-500 metal of enemy units, more if the units are particularly fast. This means that, no matter what you build, you'll probably be able to kill an exposed commander. There are certainly units that are better at it than others, but few are hard-countered by commanders. There are also numerous efficient ways to snipe commanders, such as with bombers or bursty artillery. This is interesting but not without issues.
Commanders are the heaviest unit on the field from the start of the game to some way into the midgame. Their existence boosts the viability of anti-heavy units early on, which warps the viability of units at the start of the game, but in a much nicer way because it causes people to take risks with their openings. Strategies specialised for commander killing have weaknesses in other areas, which creates opportunities for counterplay. The main problem with this is that the spectre of commander sniping limits the power of all anti-heavy options that people might have in the early game, which, given flat balance, is most of them. This in turn limits the power of other, heavier units, which has effects that ripple throughout the whole game.
Alongside removing the DGun and nuke explosion we removed assassination as the default victory condition. Players could now play on after their commander died, which might have been the most controversial change because it greatly weakens the idea of the commander as the player's avatar. We did this for two reasons: it seemed like a degenerate victory condition in team games and we wanted to try playing post-commander death in 1v1.
Team commends, where a whole team loses if they lose all their commanders rather than eliminating players one by one, seemed like the only viable way to do assassination victory. We considered a lineage mode where everything that can trace its construction back a commander explodes if that commander dies, which would at least make the mechanic player ownership-agnostic. However, lineage mode could still be worked around with a lot of fiddling and fighting the UI. Players would be incentivised to track which constructors were the descendants of which commanders and trade constructors to allow some commanders to be more expendable than others.
The problem with team commends is that it is very hard to pull off against a team that is paying attention. Complete Annihilation even had extra defensive tools, such as shields, cloak fields and terraform. Any team down to just one commander would bury it deep in a well-defended hole, making the whole victory condition pointless. So it was more of a theoretical victory condition, which already weakened the idea of the commander as a player avatar.
In 1v1, assassination victory looks like a comeback mechanic, but often it just meant that the winning player would kill the underdog much more slowly and carefully, making sure to build up defences to thwart any attempt to snipe their commander. This does not mean assassination victory was useless, far from it, but it was more relevant in even games as a threat against using your commander too aggressively. This was an exciting push-your-luck mechanic, but as 1v1 players we wondered what would happen if those games could continue past the death of the commander. Plus it was not uncommon for opposing commanders to meet and kill each other, which required special rules to determine who wins. DGunning the enemy resulted in a loss while lasering them was considered fine.
We replaced assassination with complete annihilation: you win by destroying all enemy units and structures. Killing the enemy commander turned into a soft victory condition as doing so was still quite a large swing, preserving much of the push-your-luck aspect of commander use. The game could still go on though, depending on when the commander is lost, as players have more to help them recover later in the game. Early snipes are often game-ending while a mid-game commander death is just a small permanent setback that triggers a scramble over a nice chunk of metal to be reclaimed.
Later on, years into Zero-K, we moved some of the innate commander income to the aethereal realm (i.e. tied to nothing) and moved innate metal and energy storage to the commander. This increased the short-term impact of a commander loss, as it costs resources to build a replacement storage structure, and reduced the long-term penalty. The change also made storage easier to handle as the total storage available to a team should not depend on which players are on it. Aethereal income, which is technically based on the number of commanders that spawn for a team, is easy to split among its active players, while dynamically redistributing aethereal storage would be a mess.
At one point we considered removing commanders altogether. Instead of a commander, players would start with a handful of constructors. This was only ever an experimental game option though as we found that bare unarmed constructors were just too exposed. Plus, it turns out that starting with a single unique unit is still pretty cool, even if some aspects of its avatar nature had been eroded away. There have also been proposals to start players with a factory or a small base, rather than a facplop, but the aesthetic of starting with a single unit that builds everything is just too compelling to throw away.
By the time Complete Annihilation gave way to Zero-K I was quite happy with commanders. They were a neat way to start the game and had a clear role in the way they could be used early on to push out onto the map. Their presence motivated risky anti-heavy unit strategies but overall avoided warping the counter systems we had built up. It was time to just expand and balance the unit roster to shower players with options. Then in late 2010, the modular commander upgrade system was implemented, the results of which would take one or more additional posts to examine.