There are basically two big things in Ethereum's scaling roadmap. One is sharding, so each node only has to process a fraction of all transactions. Initially there will be 64 shards, but it's designed to allow up to 1024. As individual nodes get more powerful, both the number of shards and the amount processed by each shard increases, for quadratic scaling.

The other part, which is on Ethereum mainnet already, is a layer-2 idea called rollups, which store transactions on chain in a very compressed format without losing security guarantees. There are several rollup systems, capable on today's Ethereum of doing 1000 to 9000 simple tx/sec.

Once both systems are live, total capacity will be 20K to 100K tx/sec, not counting the quadratic improvements.

On the research side, there's also work to make data validation more efficient by replacing merkle trees with something more compact, like polynomial commitments. That would add another 10X factor to rollup scaling.

Regarding footguns, people are working on more rigorous languages than Solidity that still compile to the EVM. So far their compilers aren't as solid so they don't get much production use yet.

Thanks for the summary of the current state of the art -- last I looked at this sharding/federated-chain stuff was still only research grade for bitcoin as a whole. If Ethereum is ready to productize that's amazing news.

Anything more than ~3000 tx/sec will be a game changer and maybe ethereum will finally deliver on the promise of usability as currency.

At least 1000 transactions per second if not more. In terms of language security, it is possible to use a verified subset of the language in order to avoid bugs, as you can with any Turing-complete language. Few currently do, however.

Thanks -- yeah the language thing I guess is much harder to change, but I can protect from that by just sticking to relatively simple transactions, or maybe if some sort of professional review/audit agency would spring up...