Legora's take: there is no best model — lawyers write evals and Legora BAR picks the winner
soleio · x · 2026-09-18
Legora cofounder MaxJunestrand pushes back on the "which model is best" noise: the only question that matters is what produces the best outcome on the legal task in front of you.
Key points:
- The model is one layer of the system, not the system; Legora optimizes end-to-end legal outcomes.
- Models are uneven and the frontier shifts weekly — some plan long jobs, others run deep analysis across thousands of documents, and all break differently.
- Legora's lawyers write evals; every model runs the Legora BAR agentic-reasoning benchmark, and the winning model gets the work.
- The company post-trains only when it demonstrably improves customer performance.
More from coding & agent
- Dev builds LLM-free chatbot wired to tools, sparking debate on what LLMs add beyond speed — jasonkneen · 2026-09-18
- Anthropic engineer: 90% of our engineers already use self-improving loops as prompting era ends — Roger_M_Taylor · 2026-09-18
- AI Cookbook: free copy-paste code snippets for building working AI systems — Roger_M_Taylor · 2026-09-18
- 3 GitHub tools that let your AI agent scrape data from almost any website — Roger_M_Taylor · 2026-09-18
- Unsafe agent handoffs spike harm from under 5% to 95% in multi-agent systems, paper finds — dair_ai · 2026-09-18
- Jack Dorsey open-sources a framework for fully agent-run businesses, 29k GitHub stars — Roger_M_Taylor · 2026-09-18