Cohere Labs: bottom-up tool descriptions reveal three work categories taxonomies miss
Cohere_Labs · x · 2026-09-30
Cohere Labs published findings from a bottom-up analysis of tool descriptions, identifying three categories beyond what existing taxonomies of work capture. The largest is existing human work described at finer grain — where occupational databases list a single task like "maintain configuration control," real-world tools split reading, writing, validating, and rolling back configs into separate tools. The study offers insight into the real granularity of agent tool ecosystems and labor tasks.
Related event: Cohere study finds only 2.6% of agent tools map to human job tasks(4 posts)→
More from Research
- Hillel Wayne: TLA+ is great, but 'formal methods will save AI' hype misses what it can't even express — leland_mcinnes · 2026-09-30
- Frontier Data Summit showcases a dozen new AI benchmarks and a push to rethink evaluation science — sanmikoyejo · 2026-09-30
- SwitchSD, speculative decoding that switches between neural drafting and context copying, accepted to NeurIPS — arankomatsuzaki · 2026-09-30
- Decision model Jev makes web agents 4.3x faster and 13x cheaper across 240 runs — EdenEmarco177 · 2026-09-30
- AI protein design startup Profluent Bio partners with BioMarin — nathanbenaich · 2026-09-30
- Eric Topol highlights gut microbiome patterns predicting early-onset cancer outcomes — EricTopol · 2026-09-30