from 59% to 94%
agent retrieval accuracy after improving descriptions and search.
New Research · 2026
Discoverable by Design
How API description quality drives AI agent performance, and why better data beats a bigger model.
AI agents live or die by whether they can find the right tool at runtime. New research from Jentic, benchmarking 8,467 operations across 70 real enterprise APIs, shows that improving API description quality and search lifts agent retrieval accuracy from 59% to 94%.
By the Numbers
What we found across 8,467 API operations
from 85% to 91.5%
agent search success from improving descriptions alone.
around 80%
model-cost reduction using a smaller model, with no loss in selection accuracy.
43%
of catalogued API operations have both a summary and a description. Most don’t.
From the authors
Watch the video from the authors
Adam Bermingham (Head of Data) and Emma Murphy (Software Engineering Intern) introduce Jentic's research on why agents struggle to find the right tool, and how better API descriptions fix it.
What you'll find inside
Better data beats a bigger model
Most teams try to fix agent reliability with a more powerful model. The data points somewhere cheaper: the descriptions your agents actually read.
Why tool discovery, not tool invocation, is what makes or breaks agent reliability.
How improving API descriptions and search lifts retrieval from 59% to 94%, and why targeted fixes to your most critical APIs work almost as well as fixing everything.
Why a cheaper model on good data beats an expensive model on bad data, cutting model cost ~80% with no loss in accuracy.
Ready to learn more?
Get your copy of the whitepaper
Fill in the short form and we'll send it straight to your inbox. No spam — ever.