You should specify what sort of environment you're talking about. Most of the methods I've heard of also use batteries to store the energy, but you could also use an array of supercapacitors. However, that's getting expensive as well.
Ideas I've seen used successfully to some degree:
Solar is the first one, but it does require light. Any light will do of course, even fluorescents. Will it keep your circuit powered for a significant amount of time? Depends on where it's put.
Vibration/Piezoelectric generation has been used successfully to power sensors placed under busy stairs. The key is busy stairs - imagine the New York subway or the main stairs at a university between classes.
EM capturing - If there's a lot of ambient EM then you can put out an antenna, rectify the results and regulate it - boom, power. However, it's usually illegal to do this with significant sources. For example, you can power a light bulb if you're with a quarter mile of a large radio antenna, but harvesting that power is illegal in most countries. It WILL be noticed too.
As Endolith said, you'll probably want a less power-hungry module. Consider Zigbee instead of WiFi - it was designed to be used in low-power devices that transmit for short bursts then stop.
[EDIT] Ok, office/home/domestic is the application. If you control it significantly, maybe you can alter it. Say by adding inductive chargers? I don't think there's that many source of power in a standard office other than solar.
Digi has XBee modules which have the ZigBee protocol built-in and present themselves to your AVR as a simple UART, easy to interface. Digi has point-to-multipoint and mesh solutions. Modules are compact and reasonably priced.
Running off a coin cell may not be possible; RF transceivers are rather power-hungry (10s of milliAmps). Unless you have a very small power duty cycle. If for instance the device is switched on for 1 second every ten minutes you have a duty cycle of 0.17%.
I use wireless electricity everyday.
In my toothbrush:
And in my cell phone:
The method used in my devices is called Inductive Charging. I talk a bit more about it in my answer to this question. This is the most common and most practical form of transmitting energy wirelessly at the moment. But as many of the comments have noted, this is considered near field transmission. And with an effective range of only a few millimeters, it is very near field.
The amount of energy transferred and the efficiency of the transfer can be increased quite a bit (although still considered to be near field) by adding a capacitor to each of the inductor coils and tuning the resultant RLC networks to have a high Q factor at the same (resonant) frequency. A team from MIT did research into using inductive resonance as a wireless power transfer system.
The researchers have since formed a company called WiTricity to further develop the technology. While they still haven't brought a product to the commercial market, they have made some impressive demonstrations:
It's important to note that the distance between the transmitter and the receiver plays a crucial factor in determining how much energy can reliably be transferred. As can be seen in this paper based on the MIT project, the decay in voltage with respect to distance between the coils is exponential:
But there are many other methods such as microwave and laser that are capable of much greater distances. However, these methods are very directional and so are applicable over a much smaller area than Tesla's proposed Wardenclyffe Tower which would be omnidirectional. There are also many other factors to consider when implementing one of these methods:
And of course there is the "disturbed charge of ground and air" method used by Tesla. As far as the Tesla system goes, that got shut down because funding ran out and the stock market crashed. As for why it's not been tried since, it's primarily because such a system could not be strictly metered. Therefore, the power companies could not charge per usage and make lots of money. Without a way to monetize the technology, no investment into research and development will ever be made. That's the (conspiracy) theory, anyways. Although there are many other reasons why this method is either unfeasible or just outright wouldn't work.
I couldn't find an article with definitive numbers as to efficiency. But it's my guess that efficiency is the main reason you do not see this technology in more wide spread usage. However, it does exist, people like me (read: not rich) do have access to it, and it works quite well.
I found a case study done by the Wireless Power Consortium, makers of qi charger for my phone, which states (emphasis mine):
So the wired part of their system has an efficiency of 72% and the wireless part has an efficiency of 50%. That is using an inductive method where the coils are a few millimeters apart. Compare that to the WiTricity from Joel which states an efficiency of 40% over 2 meters.
Factor in the additional costs associated with the extra circuity and components for a wireless system as compared to the cost of a length of copper wire and you can see why long distance wireless energy transfer is still considered impractical for mass market use.