Our findings, benchmarks, and experiments on measuring & optimizing agents.https://mahalsystems.ai/https://mahalsystems.ai/research/what-survives-a-model-swap/https://mahalsystems.ai/research/what-survives-a-model-swap/Every model you build on will eventually face retirement. When that happens, don't just take the lab's successor: our six agent benchmarks show it was the worse choice as often as the better one, and much of what you tuned within your agent will have to be tuned again.Wed, 28 Oct 2026 00:00:00 GMThttps://mahalsystems.ai/research/introducing-the-aot-library/https://mahalsystems.ai/research/introducing-the-aot-library/60 vetted ways to make an agent better, cheaper or faster, across 10 categories, with the evidence for each.Fri, 23 Oct 2026 00:00:00 GMT