Engineers lacked visibility into a slow AI agent. Rebuilt low-level tracing exposed critical bottlenecks and accelerated LLM responses.
A customer service platform operates a complex distributed AI agent, with hundreds of engineers across 20 teams shipping over 100 code changes daily.
Customers began complaining that the AI agent was too slow to respond, but engineers lacked an end-to-end signal to measure actual user wait times....
“When Intercom embarked on building Fin, our main goal was nothing less than to create a truly revolutionary customer service agent.”
Intercom is a SaaS company that specializes in customer communication and engagement.
Observability platform for debugging and understanding production systems.
Related implementations across industries and use cases
Tracking spend for 300M AI agent runs was a black box. Real-time tracing now lets finance pinpoint costs and update pricing within hours.
Complex queries bottlenecked access to insights. Agents now translate plain language into expert dashboards and root cause analysis.
A bug sat for years because the fix meant a month of digging. AI traced the fragmented code to draft a solution in three days.
Tracking spend for 300M AI agent runs was a black box. Real-time tracing now lets finance pinpoint costs and update pricing within hours.
Serial testing bottlenecked development. Now, parallelized checks validate hundreds of complex conversation paths in seconds.
Moderation couldn't keep pace with 600M users. AI agents now filter toxicity while models recognize 2.5B objects to refine search.
Sellers lost hours to manual research. AI agents now prioritize leads and draft briefs, cutting prep time by 80%.
Analyzing 10TB of weekly telemetry took IT specialists days. Now, engineers ask AI in natural language to instantly retrieve charts.
Custom hardware bottlenecked imaging. A switch to software-defined GPUs now renders photorealistic 3D hearts in real time.