I've just written more:
--
Guys most of you think in too generalized manner, you attempt to predict too far future, you almost admit defeat in front of ASICs... It is a bad way methodologically.

We are living in the current iteration. There is a week passed after last hard fork to cn/r. First, it must be said, there is no FPGA or quick understanding how to program FPGA against cn/r. Technically it is possible in few hours or days. But we observe low hashrate still, week passed. So they CAN'T do that quickly at least.

Thus, fight against ASICs/FPGA is POSSIBLE. Furthermore, cn/r is quite simple algo with very basic levels of virtualization and randomization. Read my comment above, what if we introduce HARD levels of virtualization and randomization? There is physics: ASICs/FPGA can't be MUCH more profitable than CPU or GPU, if quite perfect mining algo is designed.

Today, let's focus on tasks how to add strong virtualization and randomization to the algo. RandomX with memory scratch-pad reduced from 4Gb to 256Mb (to support even Raspberry Pi) is good level of virtualization.

To add randomization we may consider external oracles like ones in Ethereum contracts or long hashes (sha512, etc) applied to concatenated transactions of last 100 blocks for example. Let's think in this direction.

Fundamental physics is on our side: if mining algo has sufficient levels of virtualization and randomization, ASIC/FPGA manufacturers MUST implement Intel-like CPU! Let them compete with Intel directly!
--
https://github.com/monero-project/meta/issues/316#issuecomment-472815812