The absolute least efficient way to get air, water, and fuel into space is the way that we currently do it: by packing as much of it as we can into rockets on Earth, and then firing it off into orbit. If this is how we have to get supplies to the moon, or Mars, it's going to be ludicrously expensive and time consuming.
A much better solution is to extract everything that we need from wherever we are: where there's ice (the moon, Mars, and asteroids all have it), there's water, air with a bit of work, and with a bit more work, rocket fuel. Plus, there are likely other valuable resources scattered around all over the place, like minerals and metals. So, great, let's get on it! But first, we've got to find the stuff. And how is NASA going to do that? Robots.
The process of robotic mining itself is well established on Earth, and NASA holds an annual Robotic Mining Competition to help drive university-level research and innovation with robots competing to mine the most simulated Martian regolith (aka dirt).
Even private companies are working on off-world mining robots. The mining hardware is a work in progress, but prospecting for the good stuff (water, ice, minerals, metals, helium-3, etc.) is more difficult. We can get a sense of generally where resources are concentrated using multispectral imaging from orbit (or from aerial platforms, where atmosphere is available), but finding the best specific little spot to start digging requires exploration and sensing at a much finer resolution.
In order to figure out the best way to do this, NASA is taking inspiration from some of the finest natural engineers on Earth that aren't beavers. Or termites. Well, they're decent engineers, I guess: ants. Or even if they're not the finest natural engineers on Earth, they're great at using the fact that there are a whooole bunch of them (something in the 300 quadrillion range) to locate and exploit sources of food in their environment.
NASA's "Swarmies" robots are designed and programmed to forage like ants do. Each individual robot has basic hardware and follows a simple set of rules, and when it finds something interesting (a barcode on the ground, in this case, but you can slap whatever sensor package on them that you want), the robot calls over all its Swarmie friends to come help it out.
The current incarnation of this system only uses four robots, but it's been designed with scalability in mind, and it'll work for all different kinds of hardware. All of the benefits of swarm robotics apply here: you can deploy lots and lots of small, cheap robots that can work together to efficiently perform much of the work that would take one big, expensive robot a very long time to execute. And if you lose one or two individuals out of your swarm to mechanical issues, moonquakes, Marsnadoes, asteroid impacts, meteor impacts, meteoroid impacts, meteorite impacts, rogue AI takeovers, alien abductions, climate change, budget cuts, untrustworthy Nigerian princes, or an attack on the Deep Space Network by crazed wombats, the mission will still likely succeed. Hooray!
Next, NASA will add some robots to the mix that actually do know how to get some work done. The plan is to incorporate RASSOR, "a concept robotic vehicle evaluating designs for a future craft that could work on another world."
[ NASA ]