Berkeley & AWS: agents return faster code with Lean proofs of equivalence, 1.85x mean speedup
xiuyu_l · x · 2026-10-24
UC Berkeley and AWS introduce Agentic Verified Optimization: an agent receives an existing function and returns a faster version plus a Lean proof that it equals the original on every input. Existing code serves as the spec, so no manual formalization is needed; speedups only count with a proof, making reward hacking hard.
- On Verina's 189 reference implementations, agents produced faster proved rewrites for 156, with a 1.85x geometric mean speedup
- Their Aqueduct tool migrates C, C++, Go, and Rust into native Lean: across 17 migrated apps, 92 proved rewrites sped up 47 of 51 scenarios, and 15 optimized functions beat the matched native operation
- Key finding: application-level gains (1.12x) are far smaller than function-level gains (1.86x), making function selection a core problem
More from coding & agent
- OpenAI DevEx lead to keynote BoxWorks Dev Summit on tools, memory and agentic work — OpenAIDevs · 2026-10-24
- Developer slams Kimi coding plan: 5-hour quota gone in under 10 minutes — jasonkneen · 2026-10-24
- Terminal-Bench 4.0 chart: Kimi K3.1 Pro tops at ~65% for ~$4-5 per task, beating Claude on cost — zainhas · 2026-10-24
- Dev builds a bookmarklet that clones your clipboard into a reusable paste bookmark — JnBrymn · 2026-10-24
- ZXFLASH compiles ZX Spectrum BASIC straight to Flash SWF bytecode running in Ruffle — danbri · 2026-10-24
- Enclave opens autonomous security agent fleet to everyone, launches Threat Feed vulnerability feed — BenBajarin · 2026-10-24