Warning: You seem to assume there's "one" architecture for OFDM. That's not the case.
Note that this is all things I've seen or can very well imagine happening in real-world systems – there's no "one correct OFDM system", which all other systems are identical to. Reality is often very complex, and the fact that you often use OFDM on wideband, high-SNR channels means that you'll also want to use QAM on the subcarriers, in which case the receiver can't just adjust gain arbitrarily without adjusting its notion of amplitude, so I'm really not sure I'd call a gain adjuster in an OFDM receiver "AGC" – AGCs are typically power locked loops, and that just won't work if the receiver afterwards is a multicarrier system.
bits - symbols - IFFT+CP - precoding - DAC - filter - PA - up conversion
typically, you power-amplify the RF signal, not the baseband signal, so
bits - symbols - IFFT+CP - DAC - filter - up conversion - bandpass filter - PA - bandpass filter - antenna
makes more sense. When you've got a PA, you almost also have bandpass filtering around that, to keep harmonics due to nonlinearities out of your band – especially important in OFDM, which has a large PAPR/dynamic range.
(I've also removed the "precoding" part on your transmitter, there's no matching thing on your receiver, so how would the transmitter know how to precode?)
On RX, you definitely apply an LNA on the RF signal before it goes through any other active elements like the mixer:
Antenna - band pass filter - LNA - mixer - low-pass filter - ADC - Remove CP + FFT - Equalizer - symbols - bits
where does the AGC sit here
Nowhere. The thing classically called "AGC" is an amplifier sitting in a control loop that observes the amplifier output's power. That won't work for a multicarrier system with amplitude-sensitive modulations on the subcarriers. (You can of course have such a classical AGC, if it's very slow to adapt, if you have a very stationary set up, so that your gain can be adapted based upon observing many OFDM symbols over one unchanged channel. The one use case that comes to mind would be fixed digital TV reception. But then, the AGC sits "before anything else", i.e. after the antenna in my diagram, and "left of" your diagrams completely.)
When we want to call something "AGC", then that'd be a behaviour of your overall system, instead of a "discrete" component.
In fact, when you sketch your system this simply, there's plain no AGC necessary at all: your equalizer does exactly that.
Now, in reality, receiver amplifiers aren't noise free and perfectly linear over a huge dynamic range, so having adjustable gain elements does make sense.
Often, you'll find systems that look more like:
Antenna - band pass filter - optional adjustable attenuation - LNA - further adjustable amplifiers/attenuators - mixer - low-pass filter - ADC - Remove CP + FFT - Equalizer - symbols - bits
and then the gain of the adjustable stage(s) is adjusted according to the observation of the symbols, typically when looking for preambles or the position of the CP; but the synchronization algorithm isn't part of a simplified "fixed" flow graph like yours, so it "sits" nowhere in that. But think about that: for knowing when to cut off the CP, you need to detect the CP. The detection of the CP will usually also yield an estimate for the signal power, which you can use to put your amplifiers in their optimal linearity region.
You mention "precoding" in your original TX, but that requires knowledge of the channel at the transmitter. When the transmitter knows the channel, then chances are your receiver doesn't need any self-adjusting gain: the transmitter would just send the power needed to keep your receiver at its optimal operational point.