A programmer showed that large language models can automate proof checking for a Zstandard decompressor written in Lean, a dependently‑typed programming language. Dependently‑typed languages such as Lean let developers encode precise invariants, but constructing proofs can require many times the effort spent on design and implementation, as seen in the seL4 effort where proof work exceeded implementation work tenfold. The demonstration built an FSE entropy encoder following the Zstandard specification and proved formal properties about table size, state distribution, and reachability. Large language models completed the proofs in under an hour using a modest subscription, though the proofs required adjustments to avoid imperative code patterns that hinder verification. The author noted that the type checker can consume large amounts of memory for complex proofs and that maintaining proofs as code changes remains a challenge. Small experiments with verified AArch64 assembly showed equivalence for tiny functions but scaling proved difficult. The work suggests that LLMs may lower the barrier to using dependently‑typed languages for practical software, but broader adoption will depend on tooling maturity and proof‑maintenance costs.