Claude 3.5 Sonnet

Block uses Claude through Databricks in its open-source goose agent to generate SQL, query datasets and connect internal tools. The company reports adoption across 15 job profiles. Reported engineering time savings concern goose's broader use and do not isolate natural-language data access.

AppFolio's Realm-X Messages uses Claude to draft property-management replies using resident history, policies and maintenance context. Beta customers reported saving about 11 hours weekly. The source also reports 26 seconds saved per message; the weekly figure is customer-reported.

Lokalise integrates Claude into its translation orchestration and editor, where translators accept, revise or reject suggestions. An internal A/B test reported an 82.6% acceptance rate for Claude 3.5 Sonnet. Broader 80% cost-saving claims concern Lokalise's multi-model platform.

NRI combines Claude, review checklists and previous examples to identify issues in specialized Japanese business documents. Its system filters insignificant suggested edits. The company reports a 50% reduction in review time; detailed evaluation data and code are not published in the story.

Quillit uses Claude to summarize interviews, query qualitative datasets and attach references to transcripts or recordings. The vendor claims up to 80% less report-writing time. It reports internal citation accuracy of 89% to 98% with Claude 3.5, so references still require checking.

Law&Company uses Claude with retrieval of legal sources in SuperLawyer for research, precedent analysis, drafting and document questions. In a user survey, 92.5% reported saving time. These are self-reported results rather than independently measured lawyer productivity.

Thomson Reuters uses Claude through Amazon Bedrock within CoCounsel to synthesize tax and legal reference material using retrieval. Expert validation supports its quality process. The story includes customer expectations of time savings, rather than a measured tax-research productivity result.

Orange uses Claude in a manga-localization pipeline combining translation, image processing and professional editing. The company reports generation time dropping from more than ten hours to under two per volume. It says translators retain 50% to 75% of Claude's proposed translations.