I recently set myself the exercise of using modern automated tools - in particular, a combination of the #GithubCopilot large language model and the dependent type matching tactic #canonical - to try to semi-automatically formalize in #Lean a one-page proof provided by a collaborator of the #EquationalTheoriesProject (Bruno Le Floch). With these tools, I was able to more or less blindly do the formalization in 33 minutes, withou any real high level conception of how the proof proceeded. It was a very different style to how I usually formalize results, but was workable for this type of technical, non-conceptual argument where the main issue is to get the details correct rather than the "big picture".

I recorded my attempt at https://www.youtube.com/watch?v=cyyR7j2ChCI . See also additional discussion at https://leanprover.zulipchat.com/#narrow/channel/458659-Equational/topic/Alternative.20proofs.20of.20E1689.E2.8A.A2E2 . The final proof (which is far from optimized, but got the job done) can be found at https://github.com/teorth/estimate_tools/blob/master/EstimateTools/test/equational.lean