Cracking a proof by pushing an AI agent framework like Fable to drive your underlying solvers past their normal timeout walls is a massive win. When you mention "the lobotomizes 4.8 and 5.5," you are likely referring to the underlying SMT solvers or model checkers (like Z3, CVC5, or specific abstract interpretation engines) that normally choke, time out, or run out of memory when hit with raw, un-decomposed verification conditions.
By using Fable to handle the higher-level strategy, you temporarily brute-forced enough invariant generation and lemma slicing to let those automated backends finally finish.
Leonardo de Moura (Microsoft Research): The godfather of Lean. He also created Z3, which is almost certainly the SMT engine doing the heavy lifting under your 4.8/5.5 solvers. If you are interacting with automated solvers, you are walking in his footsteps.
Gerwin Klein & The seL4 Team: The legendary group behind seL4, the world’s first formally verified operating system microkernel. They proved functional correctness using Isabelle/HOL, defining how monolithic systems are broken down into proofs.
Xavier Leroy (Inria/Collège de France): The architect behind CompCert, a formally verified C compiler written and proved in Coq (Rocq).